@rig-c/pose

Puts motion from a pose onto a rig. The motion is joint positions over time: from a video, a dataset, or points placed by hand. The rig is one rig-c built. A body-18 pose sequence becomes bone keys in rig-c's motion format, and the keys are checked before anything is written.

0.x is the prototype. The format, the command and the output can change in any release until 1.0.

Part of the rig-c toolchain: core compiles and gates a rig, parts builds one from a painting, play plays it, and pose moves it.

Try it

examples/wave is in the repository, not the npm package: clone the repository, or copy its four files. It runs with nothing else: a bones-only stick rig (no art), a 3-second wave made from joint angles (make-pose.ts writes it), and the map that joins them.

rig-pose retarget --rig examples/wave/rig --map examples/wave/map.json --pose examples/wave/pose.json --out wave.motion.json --build wave-build

pose.json is already cleaned (make-pose.ts passes it through clean). It prints the gate's verdict and where every mapped bone landed, then writes wave.motion.json and the gated build. rigc preview --candidate wave-build plays it in the Spine Web Player. A rig with no art shows only its bones.

What the rig needs

The map names chains: each limb as bones that descend from one another, parent to child. That means a thigh with its shin below it, and an upper arm with its forearm below it, and likewise for the trunk and the head. A rig whose arm is one bone, or that has no leg bones, cannot take a pose: the map is refused by name (MAP_*). @rig-c/parts builds such a rig from a painting's keypoints with propose --keypoints. A rig drawn by hand needs the same chains.

Clean first: `rig-pose clean`

rig-pose clean --pose dance.scan.json --out dance.pose.json [--loop] [--legs 5] [--rest 7] [--hips-inferred] [--hold off] [--hold-speed 0.015] [--hold-reach 0.03]

retarget reads only a sequence that clean wrote (#67). Where the video's camera moves, the order is scan → clean → retarget (#80): a zoom step is a jump of the picture, and clean's spike and smoothing steps would read it as motion of the body, so tools/scan.py takes the camera out of the raw sequence first and clean reads what it writes. A pose model reads each frame alone, so a hanging wrist moves ±20 px a frame and a cloak-hidden leg jumps 100+ px, and every key carries that shake. clean takes it out once, at the sequence, where it can be inspected. A sequence without the record is refused by name (POSE_NOT_CLEANED), and the message is the clean command to run on that file. There is no flag to skip the step. Windows of 1 (--legs 1 --rest 1) are a legitimate run that smooths nothing, and they are recorded as such.

The steps, in this order, each printed with its figures:

Spikes. A joint more than 0.2 trunk lengths from its 5-frame median takes the median. The trunk is neck → mid-hip, the median over the clip. Each replacement is listed by frame and joint (SPIKE). Without --loop, a clip end cuts the window of the first and last two frames. There the median is a neighbour's value, and a joint moving steadily would read as a spike. Those frames are not tested, and the SPIKES line and the record name them. Reflecting the window does not help: the median of [x2, x1, x0, x1, x2] is still x1.
Occlusion. A joint whose state is not observed (or inferred), or whose score is under 0.3, takes its 3-frame median (OCCLUDED).
Smoothing. A quadratic Savitzky–Golay per joint: the legs (both hips, knees and ankles) over --legs frames (default 5), every other joint over --rest (default 7). A window is 1 (none) or odd and at least 5. 3 is refused because a quadratic through 3 points passes through all three. Without --loop, the window is cut at the ends. Where fewer than three frames have a position, the degree drops to fit the ones that do.
Floor. Only with --loop: the whole figure moves in y by one constant, so that the lowest ankle over the cycle is where it was before smoothing. A loop is also written with "travel": "in-place" (#92, the TRAVEL line): one cycle of a gait, repeated, does not travel, whatever its supporting foot does in the picture (below).
Hold (#91). A joint that has stopped is written at one position (HOLD), because detector noise on a still joint is not motion, and no smoothing window removes it without also smearing the movement into the stop. Below.

A median is taken per coordinate (x and y apart), over the window's frames on which the joint has a position, its own frame included, whatever its state. An even count takes the mean of the middle two. --loop says the sequence is one cycle, such as a gait from strike to strike. Every window then wraps, the last frame running into the first, so nothing is untested; a cycle shorter than its widest window is refused (CLEAN_LOOP). Nothing is invented. A joint with no position on a frame gets none: retarget holds it and prints a HOLD line. States and scores are written as they were read. --coco reads COCO-17 (below) and converts it first. --person picks the person when there is more than one, and only that person is written.

The output carries the record, here the dancer's from the table below:

"cleaned": { "spec": "rig-pose-clean/1",
  "params": { "loop": false, "legs": 5, "rest": 7 },
  "spikes": { "found": [], "untested": [0, 1, 311, 312] },
  "jitter": { "before": 7.9123, "after": 4.6225 } }

jitter is the mean |second difference| of y, in px/f², over every joint and every three consecutive frames that have a position, wrapping with --loop. It is measured on the sequence as read and on the cleaned positions. retarget prints the record as its CLEAN line and copies it into --report.

--hips-inferred (#63) declares that the hips are not seen: under a skirt or a cloak the pose model still places them, but where it puts them is its guess. After the steps, r_hip and l_hip are marked inferred on every frame where they have a position (INFERRED), and the record's params carry "hipsInferred": { "by": "declared" } — the person running clean said so. Positions and scores are unchanged, so retarget's keys are the same with the flag and without it. What changes is the POSTURE line, which does not read a hip tilt off inferred hips (below), and the CLEAN line, which ends hips inferred (declared at clean).

The stillness hold (#91, #108, #109) runs on the cleaned positions, after the smoothing and the floor, joint by joint — except the five face joints (nose, r_eye, l_eye, r_ear, l_ear), which are held as one group (all of them or none, over the span where all five pass, #109), so a piecewise hold does not turn the head. Stillness is a property of a joint (or of the face as a whole), so a hand that has stopped is held while the elbow beside it still settles. A joint's speed on a frame is its movement from the frame before, in trunk lengths, taken as a 5-frame median. A hold of a joint is a span of at least 0.5 s (the first frame's time to the end of the last) on every frame of which its median speed is under --hold-speed (default 0.015 trunk lengths a frame), and over which it stays within --hold-reach (default 0.03 trunk lengths) of its own span median. Speed alone cannot tell a slow drift from stillness, because a drift accumulates. For each joint (and for the face group together) the span is chosen greedily, left to right: from a still frame it grows one frame at a time while the reach holds on the grown span, its median read again each time, and stops at the first frame that breaks it. If what it reached lasts 0.5 s it is a hold and the search resumes after it; if not, the search moves one frame on. A still run not held by that greedy pass that lasts at least 0.5 s and fails the reach only on one interior run of at most 6 frames stays held around that excursion (#108), provided the held frames themselves last at least 0.5 s and stay within the reach of their own median: the excursion is kept (blended from its cleaned positions toward the span median over the 6 frames from either held side), not erased, and the joint is not released for the whole span. A span does not cross a loop's seam.

On every held frame of a hold the joint takes its per-coordinate median over the span. It blends in linearly over the 6 frames before it (the 6th stays as cleaned, the frame before the span is ⅚ of the way) and out over the 6 frames after it, mirrored. A blend frame inside another hold of the same joint is not touched; where two blends of one joint meet, the larger weight wins. Where one hold of a joint ends on frame k and the next starts on k + 1 + g, with a gap g of under 6 frames (a seam: the reach broke while the joint was still moving slowly, or a few frames between two holds failed a test), the two medians differ, and the joint would jump between them. A seam cross-fades over 6 + g frames, symmetric about the gap's centre: frames k − 2 … k + 3 + g take 1/(2(6 + g)), 3/(2(6 + g)), … of the way from the first median to the second (for g = 0: 1/12, 3/12, … 11/12, the blend's sixths placed symmetric about the seam), so no frame moves more than 1/(6 + g) of the step. The cross-fade replaces the blends on the gap's frames. Each seam is printed (SEAM) and recorded with the joint, the second hold's first frame, the step in trunk lengths and the gap. A hold that carries a cross-fade or an excursion has a non-zero jitter after: the ramp's or excursion's own curvature, printed as (ramp) or (excursion), not noise. A joint that fails keeps its cleaned track. A joint with no position on a frame gets none, and states and scores are kept.

Each hold is printed (HOLD, joint by joint, with its largest speed and reach, and its excursion when one was kept) and recorded with the joint's jitter inside it, before and after. Where some joints are held, every run of the same frames (the held spans joined where they overlap or touch) on which a joint has no hold is named with the test it fails there and its figure (UNHELD, and the record's unheld): speed (its largest median speed), else reach (its largest distance from its median over the run), else short (it is still, but for under 0.5 s; the figure is its seconds), else span (it passes all three, and its spans broke elsewhere).

"params": { "loop": false, "legs": 5, "rest": 7, "hold": { "speed": 0.015, "reach": 0.03, "minSeconds": 0.5, "blend": 6 } },
"holds": [{ "joint": "nose", "frames": [first, last], "excursion"?: [first, last], "jitter": { "before": before, "after": 0 } }, …],
"seams": [{ "joint": "neck", "frame": second hold's first frame, "step": trunk lengths, "gap": frames }, …],
"unheld": [{ "joint": "l_elbow", "frames": [first, last], "test": "reach", "value": figure }, …]

The CLEAN line names, for each region, the joints held over which frames and the frames not held with their test and figure and the seams cross-faded (HOLD frames a–b: held nose, neck, … a–b; seams cross-faded: neck f (step …); not held: l_elbow c–d (reach … > 0.03)), or says no HOLD. --hold off turns the step off: the positions are those of a clean without it, and the record carries "hold": "off" and none of holds, seams and unheld. A record written before #91 names none of them, and is read as it was.

Measured hips: tools/pelvis.py (#86, not shipped) reads the pelvis line where the eye reads it under a skirt, off the skirt's waistband, and runs after clean: scan → clean → pelvis → retarget. It takes the video and a sequence clean wrote without --hips-inferred. It refuses by name, writing nothing:

a sequence with no cleaned record (PELVIS_NOT_CLEANED);
one whose hips are already marked (PELVIS_HIPS_INFERRED);
a frame count other than the video's (PELVIS_FRAMES);
a clip on which no frame gives a reading (PELVIS_NO_READING).

On each frame it reads the top edge of the dark region at the waist. The region is set by the hips and the frame's trunk length. The tool keeps the edge's flattest high run, since a flaring skirt's sides are steep and are not the pelvis line, and fits a line to it.

A frame gives no reading, declined with the reason, when:

a wrist lies in front of the waist (the picture under it is not the skirt's dark);
the edge is too broken or too far from a line;
the reading is over 25° from level, or a lone outlier: further, per frame, than the misread guard allows from both the read before it and the reading after it.

The tool never draws a line through a hand in front. If two reads lie close in time on either side (at most 1 s apart), a declined frame takes the linear interpolation between them. Over the clip's unheld read frames (or all read frames when fewer than two are unheld), the tool fits a zero-intercept gain (fit.gain) and correlation (fit.corr) between the detector's hip angle and the read waistband angle (#106). On a declined run longer than 1 s (or at the clip's start or end), when fit.corr >= 0.9 the frame takes the detector's hip angle scaled by fit.gain (scaled); when fit.corr < 0.9 the detector line does not track the waistband on that clip, so the run is marked unmeasured (keeping the detector's hip pair, refused nowhere, and named on the line).

Both hips are marked inferred on every frame and turned about their midpoint to the frame's angle, keeping the midpoint and the width. The record says which frame is which: "hipsInferred": { "by": "measured", "method", "fit"?: { "frames", "gain", "corr" }, "read", "interpolated", "scaled"?: [[first, last], …], "held": [[first, last], …], "unmeasured"?: [[first, last], …], "declined": [{ "frames": [first, last], "why" }] }. Every frame is in exactly one of read, interpolated, scaled, held and unmeasured, and declined lists exactly the non-read frames. Runs are frame indices inside the sequence, increasing and disjoint. A malformed record is refused by name (POSE_CLEANED).

tools/pelvis_selftest.py is its synthetic check. A planted wrist on a hand in front of the waist reads "covered", and with the hand moved away the edge's angle is read again. The misread guard, the interpolation, the detector-gain scaling on long covered spans (#106, with --hold-covered as its planted mutant) and the unmeasured path (r < 0.9) are checked frame by frame. Its cover mutant (--no-cover-check) must read a line through the hand, and the check asserts that it does.

Measured on HQ's references (#67's bars):

Clip	Run	Jitter, px/f²	Against
The 13 s dancer	defaults	7.91 → 4.62	HQ's hand-cleaned file: 4.62; positions within 0.071 px
armor_walk, one 32-frame cycle	`--loop`	9.43 → 3.16; 3.155 without the four arm joints	HQ's file without the arm joints: 3.156 (HQ's hand step smoothed the arms a second time)
armor_run, one 20-frame cycle	`--loop --rest 5`	17.06 → 7.99	HQ's file: 8.17
school_walk, 90 frames, held out	defaults	5.99 → 1.95 (−67 %)	the bar: −30 %
school_run, 90 frames, held out	defaults	47.00 → 11.54 (−75 %)	the bar: −30 %

A retarget of the cleaned dancer, with --flip 90 --short 0.35 --bends clamp --leg-scale 0.8,1.05 --plant-feet, finds the same jumps (FEET) as the same run on HQ's hand-cleaned file. Its hip rotate sd over frames 94–175 is 4.574, against HQ's 4.58.

What it does

rig-pose retarget --rig out/rig --map map.json --pose dance.pose.json --out motion.json --name dance

For every frame of the pose and every bone the map names, the segment that bone carries (an upper arm: shoulder → elbow) points somewhere in the picture. The bone is turned so its segment points the same way: the picture angle, flipped into rig-c's y-up world, minus the bone's angle at the rig's setup pose (read through rig-c's own poser), less the turn of the mapped bone above it — because a rotate key in rig-c adds to the setup rotation and a child turns with its parent. The travel of the pelvis midpoint, scaled by a length measured on both the rig and the pose, moves the hip. When a head chain carries the eye line (r_eye → l_eye), the eyes' vertical travel against neck in the reference, scaled into the rig's units, keys a vertical translate nod channel on the head bone (#111), leaving its rotate (roll) track unchanged. The result is one animation in rig-c's motion format (rigc-motion/1): a rotate track per mapped bone, a translate track on the hip (and on the head when the reference nods), one key per pose frame — and, when the map lists alternate drawings, an attachment track per slot they switch (and a translate track per bone a set offsets).

Nothing is written until every check passes:

Check	Fails with
rig-c's gate is green with the rig as it is, and again with the animation in (`rigc build`)	`RIG_GATE_BASE`, `RIG_GATE`, then rigc's own `FAIL` lines
every mapped bone lands on the angle the pose asks for, on every frame — the gated build is posed by rig-c's poser and measured (0.01° / 0.01 units); with `--leg-scale`, every scaled bone's x axis on its asked scale (0.0001 of its setup length), and with either leg flag each ankle where the keys put it (0.01 units + 0.0001 × its setup distance from the hip)	`POSE_LANDS`: the bone, the worst error, the frame
elbows and knees bend only the way they can: at every joint inside a chain, the inner angle stays on the side of the rest pose's	`POSE_BEND_ONE_WAY`: the chain, the joint, the frame, the angle
every point the map needs is good on at least one frame	`POSE_POINT_NEVER_GOOD`
the map resolves against the rig: bones exist, chains descend, aims descend	`MAP_*`, naming both sides

and what the run cannot decide for you it lists:

RIGC — the rig-c version and entry that gated and posed the run (also rigc in the report): a later rig-c can change the bytes.
HOLD — a joint missing, or scored under --min-score, holds its last good value; one line per span, with the frames and the reason.
SHORT — a segment shorter in the picture than --short of its median length (with --short <fraction>; off by default) has its angle there interpolated, not read; one line per span, with the chain, the segment, the frames and the lowest fraction. Below.
END-ON — a SHORT run lasting at least 0.15 s: the limb points at the camera, a pose the rig cannot draw.
FLIP — a segment that turns more than --flip degrees in one frame (with --flip <deg>; off by default) is spread over the frames around the jump; one line per spread, with the chain, the segment, the jump, the frames and h. Below.
BEND — with --bends clamp, a hinge bent the wrong way is straightened on that frame instead of refusing.
FOLD — rigc render --folds over the new animation: every mesh that folds, and when. A reading, not a gate.
POSTURE — when the map has a pelvis chain (joints r_hip, l_hip) and a translate: the hip line's tilt range, and the sway — the x-range of mid_hip − mid_shoulder — in rig units and in shoulder widths, the pose's beside the posed rig's. A reading, not a gate. A frame on which either hip is inferred is left out of the tilt. When that is every frame, as after clean --hips-inferred, the line is hip tilt not read: hips inferred (declared at clean) and the report's posture has "tilt": null with "tiltWhy"; on some frames only, the tilt is read over the others and the line counts the frames left out (posture.hipsInferred). The source is read from the clean record, or is source not stated when the joints carry the state and the record does not. With a measured record (#86, #106), the read, interpolated, scaled and held frames are the measurement, so the tilt is read on all four, while unmeasured frames (fit.corr < 0.9) are left out and named. The line adds tilt measured (<method>) on N frame(s): R read, I interpolated, H held (held: <spans>) (plus scaled and unmeasured spans when present). The report's posture carries tiltMeasured. The sway reads the hips as before. With --plant-feet, POSTURE also prints one bounce line per support span (#107, posture.bounce in the report): the detrended (centred ±12-frame moving mean removed) vertical standard deviation, rig/reference ratio and correlation for the hip, the neck and eyes − neck.
SWITCH — with alternates in the map (below): each span on which a set of alternate drawings shows, its slots, its frames, how far the pose was from the set's key skeleton, how far each bone with a drawn angle was turned from it, and where each of its boundaries was moved to (moved, below).
UNCOVERED — with alternates in the map: every fold on a frame where no switch took the folding mesh off screen. This is the list of poses still to paint.
RANGE — a chain's range (below): a bone whose rotate key the pose asks past its range, outside every span of a set depicting its chain, is clamped at the limit; one line per run, with the frames, the seconds and the largest overflow. The demand for an alternate.
ART-RANGE — inside a set's span: a bone with a drawn angle in the set, turned from it past its limits — the set's turn for that bone (held at the limit) or, without one, its chain's range (the bone follows the pose). Listed as the pose asked it either way: the demand for another drawing of that set.
RAISED — with --raised-by-head (below): one line per span of a set that draws an arm raised, per chain. It gives the frames and the median eye distances (the rig's in units, the reference's in px). It gives the wrist's offset from the head centre in eye distances: the median of the reference's, of where the wrist was placed, of the rig's on the pose's angles, and of the rig's as keyed. It gives the worst |rig − reference| before and after, the worst |after − placed|, and the frames whose upper arm the set's turn held. RAISED none when the flag is given and no such set showed.
BY-HIP — with a set's "place": "by-hip" (below): the RAISED line against the hip, one per span of such a set per chain it places. The unit is the hip width, the offset is from the mid-hip. BY-HIP none when the map has such a set and none of its spans showed.
REACH — with --raised-by-head, or a "by-hip" set: a placed span whose target was out of the arm's reach, too far or too near. It names the frames, which way, and the shortfall in eye distances (hip widths for "by-hip"). The wrist was placed at the reach limit instead. This is not ART-RANGE: the rig's proportions were the limit, not the art's range. A too-near frame is placed at the limit like a too-far one, and graded at the placed distance.
LEG-SCALE — with --leg-scale (below): per leg segment, its straight length, the spread of its picture length over it, the frames clamped at each limit, and (with --plant-feet) one past the floor (support) line per supporting run that asked under the floor (#133, legScale[].support in the report).
PIVOT — when the map translates a bone and has both legs: the translated bone's origin against its thigh roots' midpoint at setup, in units and hip widths — the pivot every hip turn swings the legs about.
FEET — when the map has both legs: the supports and jumps read off the reference ankles (planted or not), and on the posed build, outside the jumps, the lowest ankle's vertical range and each support span's worst vertical range and horizontal drift of the supporting ankle, in leg lengths.
STANCE — with --stance (below): one line per run of frames planted both (with the stretch keyed, its largest k), lifted, or past the tolerance below (the pose's height kept, #114), with the free side, the frames and the tolerance; and one per run whose stretch fell past its bounds (past the bounds), with the k the floor needed, the bound it had and how far the foot was left from it; and, where the leg's own range set a lower bound below 0.92, one bound from the leg's own range line with each side's lowest.
SHEAR — with --leg-scale: every scaled bone and every bone under one, how far its art is sheared and how long its world axes are against setup. A reading.
IDLE, IDLE-DROPPED, IDLE-KEPT-OUT, MERGE — the rig's idle against the new animation, and what --carry-idle and --merge put back (below).
REDUCE — with --reduce (below): per bone, its keys before and after and its largest error, with the frame; then the total.
FACE — with face in the map (#138): per-frame yaw and pitch read from nose against the eye midpoint (r_eye + l_eye) and, when present, the ear midpoint (r_ear + l_ear), de-rolled by the eye line, then compressed into each configured axis's [min, max] range (scale and correlation r), and written as face in --report.
TURN — a line across the body (hips, shoulders, eyes) shorter in the picture than 0.5 of its median length, or reversed, for at least 0.15 s: the body or the head turned. A reading of the pose, made whether or not the map uses the line; it changes no key. The demand for a turned-body drawing, as END-ON is for a foreshortened limb.

--report <file> writes all of it as JSON (rig-pose-report/1): spec, rigc ({ version, entry }: the rig-c that gated it), rig, animation, frames, duration, cleaned (the sequence's cleaned record, above), options (every figure the run used, switchTolerance, switchMin, switchAt, legScale, plantFeet, stance, floor and raisedByHead included), scale, holds, short, endOn, bends, flips, gate (rig-c's gate summary, a string), landings, posture, folds, legScale, pivot, feet (planted, legLength, supports, jumps, steps, changes, lowered, floor, shifts, blend, closure, and the posed reading; a jump is an inclusive [first, last] pair of frames; lowered holds { frames, was, now, cause? } runs (cause, where now is null: released or higher), present only with --plant-feet on a sequence not declared in place (#113); floor is the world y every support span was planted on and shifts holds one { side, frames, shift } per support span in world units, null and empty without --plant-feet; blend is { bounds, runs: [{ side, frames, k: { min, max } }], over: [{ side, frames, needed, left }] }, the supporting leg stretched on the blend frames, null without --plant-feet; closure is { drift, frames }, the hip's x drift closed over an in-place loop, null unless the sequence is "in-place" and --plant-feet is given), stance (tolerance, release, minSeconds, bounds, runs: { state, side, frames, maxHeight, maxDepth?, k, absorbed? } (state is both, lifted or below; maxDepth, on a below run, is the free ankle's largest distance below the supporting one in leg lengths), over: { side, frames, needed, left }; null without --stance), shear, idleDropped, idleKeptOut and merged (below), reduce (the tolerances and each row of REDUCE, below), and the demand lists. The fields a run is graded by:

field	each entry
`holds`	`{ point, from, to, source, why }`: a joint held over frames `from`–`to` at frame `source`'s value
`bends`	`{ chain, joint, frame, t, angle, side, rest }`: a hinge found bent the wrong way (`angle` in degrees, Spine's sense; `side` ±1, the side it may bend)
`endOn`	`{ chain, bone, joints, frames, seconds, minFraction }`: a limb pointing at the camera for ≥ 0.15 s
`landings`	`{ bone, worst, frame, kind, tolerance }`: where rig-c's poser put each bone, against the target. `kind` is `angle` (degrees), `origin` or `foot` (world units), or `scale` (a fraction of the setup length). A run writes only when every `worst` ≤ `tolerance`
`folds`	`{ slot, attachment, times }`: every mesh `rigc render --folds` saw fold, at those seconds; `null` when it did not read

The demand lists:

switches: [{ set, chains, slots, hides, offsets, from, to, frames: [first, last], distance: { max, median }, artTurn: { <bone>: { max, median } }, moved: { on, off } }] — one per span shown; from is the first frame's time, to the time the base art comes back (the animation's end when the span runs to it), distance in degrees over the frames shown, artTurn |wrap(target − drawn)| in degrees per bone the set has a drawn angle for. moved.on is the boundary the span starts on (null from the first frame), moved.off the one the base art or the next set comes back on (null to the end), each { frame: [selected, keyed], frames, seconds, window: [first, last], speed: { old, new } }: signed frames and seconds, the frames it could move to, and the speed in degrees per frame on the old and the new frame. A boundary between two sets is the same entry in both.
uncovered: [{ slot, attachment, chains, times }] — every entry of folds on the times its slot still shows that mesh, with the map's chains that move it (the deepest: the arm, not the trunk it hangs from). rigc render --folds reads only what the pose shows, so a mesh swapped out or hidden by a switch is not read on those frames at all; uncovered is the fold times less the switched ones.
endOn — above, under the short hold.
turns: [{ line, joints, frames, seconds, minFraction, reversed }] — line is hip, shoulder or eye.
range: [{ chain, bone, frames, seconds, maxOverflow, key, limits }] — key is the rotate key the pose asked for on the frame of the largest overflow, before the clamp.
artRange: [{ set, chain, bone, frames, seconds, maxOverflow, turn, limits, clamped, shortfall?, anchor? }] — turn is the signed art turn the pose asked on that frame; clamped says whether the set's turn held the bone there. shortfall (with --raised-by-head, or a "by-hip" set) marks a held upper arm whose re-aimed forearm left the wrist short of its placed point: the most, in eye distances (in hip widths with anchor: "hip"), with turn and maxOverflow the upper arm's on that frame.
raised: [{ set, chain, frames, seconds, eyes: { rig, reference }, reference, placed, before, after, worst: { before, after, placed }, short, upperHeld, reach: { min, max, farthest } }] — with --raised-by-head. reference, placed, before and after are the wrist's offset from the head centre on each frame of the span, [x, y] in eye distances, x right and y up. eyes holds each side's median eye distance over the span. worst.before and worst.after are the largest |rig − reference| in x and in y; worst.placed is the largest |after − placed|. short lists the REACH runs, each { frames, how, shortfall: { max, median } }, where how is far, near or missed. upperHeld lists the frames whose upper arm the set's turn held. reach gives |l0 − l1|, l0 + l1 and the farthest target from the shoulder, in world units.
byHip: [{ set, chain, frames, seconds, hips: { rig, reference }, reference, placed, before, after, worst: { before, after, placed }, short, upperHeld, reach: { min, max, farthest } }] — Present only when a set of the map declares place; a run without one writes no byHip key. Each span of a "place": "by-hip" set on each chain it places (#93): raised's fields against the hip, the offsets from the mid-hip in hip widths, hips each side's median hip width over the span, short in hip widths.

Runs in turns, range and artRange merge across a gap of one frame, as endOn does; seconds runs from the first frame's key time to the end of the last frame.

--smooth <n> is an alias kept since #67. The smoothing is clean's (--legs, --rest), so retarget prints --smooth is carried by clean; ignored and smooths nothing: --smooth 7 writes the same bytes as --smooth 1. The three paragraphs below were measured before #67, when retarget smoothed with a cubic Savitzky–Golay over --smooth frames. They are kept as the record of why the smoothing moved to the sequence.

--smooth 7 flattens the pelvis's rebound. Measured on HQ's dancer with --leg-scale 0.8,1.05 --plant-feet (#47): the rebound — the hip's peak upward acceleration at the bottom of each bounce, rig ÷ reference, median over a window, bar ≥ 0.7 — reads 0.40–0.52 with --smooth 7 in every configuration measured (0.461 and 0.501 in the two windows, where --smooth 1 reads 0.648 and 0.942). So the pelvis is read at --smooth 1, and --smooth 7 is not advised for it. What follows is the earlier advice, for the limbs.

--smooth 7 at 24 fps, for the limbs. It was measured on a 13 s dance on a @rig-c/parts rig, read back from the gated build: the motion's timing did not move (speed cross-correlation lag 0 on the wrists, ankles and hip), the frame-to-frame second difference dropped 2.5× on the wrists, 1.7–2.1× on the ankles and 1.5× on the hip, and the hip's 95th-percentile speed fell to 0.85×. The criterion set beforehand, a 2× drop on every point, was not met on the hip. That criterion was flawed: the hip's second difference is mostly real motion, while the filter cuts white noise 4.6×. A measure that isolates noise is a later card, and the default stays off until it passes.

The flip guard. A pose model can put a fast sweep of a limb into one frame — under motion blur it takes the elbow or the wrist for another point — and the rig then whips the limb round between two keys. Position smoothing cannot remove a step like that. So after holds, and before the hinge check, every segment the map aims (not a reach, which is aimed later from posed positions) is read frame to frame: a turn greater than --flip <deg> from frame k−1 to frame k is spread. For the smallest h ≥ 1 that keeps every one-frame turn at or under the threshold, frames k−h … k+h−1 are replaced by the straight line from frame k−h−1 to frame k+h, the turn's own way round. It is centred, so the step happens when the pose put it, over 2h frames, and both ends keep the pose's angles. A window that would cross the first or last frame stops there and the line says so; two windows that overlap become one. --flip off keeps every turn. The report lists each spread under flips.

The threshold was fixed by measuring first (#15). On the 13 s dance (313 frames, RTMPose), the one-frame turns of the picture angle of 10 segments — both upper arms, forearms, thighs and shins, the eye line and the hip line, 3,120 turns — run p50 1.0°, p90 5.0°, p99 23.1°, p99.5 33.4°, p99.9 78.2°, max 148.6°. 90° is the p99.9 rounded up past a margin, and marks the 3 turns above it (0.1%): both forearms between frames 181 and 182 (148.6° and 114.4°, a motion-blurred stretch) and the left forearm between 53 and 54 (99.2°, the forearm pointing at the camera). None of them is a spike that comes back: each keeps the direction the limb was already turning, and the new angle holds — a real sweep put into one frame. Holding the last angle would be wrong a frame later, so the guard spreads the step instead. With it, on the @rig-c/parts rig #11 measured, the largest one-frame turn of any guarded bone in the gated build, posed by rig-c, is 78.2° (148.6° without it), and every bone still lands.

Re-measured with the guard on, against the guarded unsmoothed run, on #11's criteria (power above 6 Hz down ≥ 2× on every point, timing lag 0, p95 speed ≥ 0.85×): no window passes. --smooth 7 comes closest — the left wrist's power drops 1.99× and the hip's p95 speed is 0.847× — so --smooth stays off by default, and so does the guard: the rule posted before the run made a default for either one depend on a window passing. Use --flip 90: with it, no bone of the gated build turns more than 90° in one frame (without it, 148.6°), and every bone still lands.

The short hold. A segment pointing at the camera is short in the picture, and its picture direction is then detector noise: the rig turns the bone to follow it. So after holds, and before the flip guard, every limb segment the map aims is measured on every frame (not a reach, and not a line across the body from a left joint to its right twin, such as the hip line or the eye line: those shorten when the body or head turns, which is a different thing, and the cutoff below was measured on limbs only) — its picture length, between the same two points its angle is read from, over its own median length across the pose. On a frame under --short <fraction> its angle is not read from the pose: it is interpolated, the short way round, from the last frame at or above the cutoff to the next one, and held flat at the first or last frame of the pose. The flip guard, the hinge check and the keys all see the held angle. --short off (the default) reads every frame. The report lists each span under short (chain, bone, segment, first and last frame, the lowest fraction).

The cutoff was fixed by measuring first (#22), over both upper arms, forearms, thighs and shins. On the 13 s dance (24 fps, 2,504 segment-frames) the fraction's 0.5th percentile is 0.51, its 5th 0.82, and 7 segment-frames (0.28%) are under 0.35; on a 16 s second sequence (30 fps, 3,936) they are 0.27 and 0.75, and 34 (0.86%). Every frame under 0.35 was looked at, and was one of two kinds:

end-on, lasting several frames — the limb really points at the camera: on the second sequence the right upper arm folded across the chest (frames 395–401, 0.09–0.25), and the right thigh and shin of a knee raised at the camera (365–370, 0.16–0.34); on the dance the left upper arm with the elbow at the camera (frames 30–37, 0.27–0.35, S2 below);
blur collapse, lasting 1–3 frames — a fast forearm whose elbow and wrist the detector both puts on the blurred hand (eleven spans on the second sequence).

Both have a meaningless direction; only the first is a pose the rig cannot draw. 0.35 is the smallest round value that takes in every end-on frame (the highest is 0.35, on the dance) without reaching normal motion (the 5th percentiles are 0.75 and 0.82). A run under the cutoff — spans merged across a gap of one frame — lasting ≥ 0.15 s (3.6 frames at 24 fps, 4.5 at 30) is the end-on kind: it excludes every blur collapse above (at most 3 frames, 0.10 s at 30 fps) and keeps every end-on span (6–8 frames). The run prints it as END-ON and the report lists it under endOn (chain, bone, segment, frames, seconds, the lowest fraction): it is the list of moments that need an alternate drawing (#6).

Re-checked on the exam01 test rig with --short 0.35 --flip 90 --bends clamp. On the second sequence the three spreads the flip guard made out of end-on segments are SHORT holds now, not FLIPs: right upper arm 399–400, right thigh 368–369, right shin 368–371. Of the flip guard's other spreads only one is left, the left forearm at 322–323; every other one fell on frames under the cutoff and is a SHORT hold instead. No guarded bone of the gated build turns more than 81.4° in one frame (85.1° with --flip 90 alone), and every bone lands. endOn there lists the right upper arm (395–401) and the right shin (365–370). It does not list the right thigh, which is under 0.35 on one frame only (369, 0.16). The plan before the run expected the thigh too, and that expectation was wrong. On the dance endOn lists exactly S2's left upper arm (30–37); the three FLIPs are #15's, the largest one-frame turn is 78.2°, and every bone lands. A first implementation also held the lines across the body. It listed the eye line (head in profile) and the hip line (body side-on) as end-on, and holding the hip line narrowed the rig's hip tilt range to 39.3° against the pose's 125.2°. The cutoff was never measured on those lines, so they are excluded; with them out the rig reads 125.2° again. The default stays off: a default needs its own criterion and a third sequence.

The legs: foreshortening and planted feet

A knee bending toward the camera is seen from the front only as a shorter leg, and a rig with rigid legs cannot lower its hip without lifting its feet (#47). Two flags, both off by default:

--leg-scale <min,max> scales each mapped leg segment (two of one side's hip, knee and ankle) along its own x axis — a scalex key — by its picture length on the frame over its straight length, the larger of the two sides' 95th-percentile picture lengths for that segment over the sequence (#148), clamped to [min, max]. The card's range is 0.8,1.05. A segment the map aims other than "x" is refused (LEGS_MAP). LEG-SCALE lists each segment's frames clamped at either limit.

rig-c's scalex multiplies the bone's x axis, and a child with the normal inherit mode takes its parent's whole world matrix. So under a thigh scaled by s, a shin at angle φ in the thigh's axes points at atan2(sin φ, s·cos φ), and is itself scaled by |(s·cos φ, sin φ)|. A child (or its intermediate control bone) with noScale or noScaleOrReflection (#153) still has its direction carried through the parent's scaled matrix, so its angle is skewed the same way, while rig-c normalises that direction to unit length so the parent's scale is not inherited. The rotate key turn − turn above assumes unit scale. A mapped bone under a scaled bone is therefore keyed through its parent's posed matrix: a rotate key k and a scalex key s' put its world x axis on its target at the scale asked (under noScale, s' is the bone's own asked scale s directly). On HQ's dancer the unit-scale formula missed the shins by up to 2.4°, their length by 0.16 of setup and the ankles by 38 units; solved, every bone lands. A normal child's art is still sheared by its parent's scale, and a bone under it (a foot) is squashed along the shin: SHEAR reads both.

--plant-feet places the hip so the supporting foot stays where it took support, through the posed legs at the scale the pose asked for (when a segment asks for a scale below --leg-scale's clamp floor, the clamp does not lift the hip by what it withheld, #107; on a supporting leg, the segment's scalex follows the asked scale down to the bound #95 derives from the leg's own range so the supporting ankle stays on #85's floor while the free leg keeps the configured clamp, and any frame clamped at #95's bound lifts the hip by the remainder so the ankle never sinks, listed as LEG-SCALE past the floor (support), #133), instead of moving it by the pose's mid_hip travel:

The supporting foot is the one whose reference ankle is lower in the picture. It changes only when the other ankle has been at least 0.02 leg lengths lower for at least 0.1 s (timed as END-ON runs are), and once confirmed takes support from the frame it crossed that margin (#102); a leg length is each side's median picture length hip → ankle, and the switch compares against their mean.
The new support is held where it is on the frame it takes over, and the hip blends in y from the old derivation to the new over 3 frames while taking the new derivation's x immediately (#102), so the planted foot does not slide across the blend.
A jump — both reference ankles and mid_hip more than 0.05 leg lengths above their rolling floor, the 95th-percentile picture y over ±1.0 s around the frame, for at least 2 frames, starting from the first immediately preceding push-off frame where both ankles are already above their rolling floor while mid_hip lags (#142) — releases the plant: the hip follows the pose's travel from where it was, and re-plants on landing (blended the same way). At a walking hand-over the front foot has not landed yet while the back ankle lifts its heel and the hip rises 0.051–0.065 leg lengths; when the front foot lands while the back ankle stays above its floor, a back ankle that holds its height in y (< 0.01 leg lengths a frame for 2 frames) is on a raised heel rather than in flight and is not read as a jump (#101). The body has to lift: an ankle's picture y drifts with depth, and under one floor for the whole sequence HQ's dancer standing still (frames 0–29 and 277–312, mid_hip moving 10 and 14 px) read as jumping, 152–155 of 313 frames in all.
In place (#92). On a sequence with "travel": "in-place", the supporting foot of a figure that does not travel moves back in the picture while the body stays. Holding it where it planted turns that slide into the body's travel, so the hip walks off and never comes back. Under the declaration each plant's anchor follows its own ankle's picture x, scaled into the rig, from the frame it planted, and keeps its y on the one floor. No step releases a foot, since a foot that follows its track has no plant to drag. Every foot may take support, and the lower one does, by the rules above. The loop is then closed: the hip's x on frame i moves by −drift · i / (n − 1), where drift is its x on the last frame less its x on the first. The last frame's hip x is then the first's, and the loop carries no net travel. The y is the floor's and is not touched. FEET prints the drift closed, in place of the steps line, and the report's feet.closure carries { drift, frames } (null otherwise). On the tests' synthetic in-place loop (40 frames, each supporting ankle sliding back 2 px a frame, the hips still), the hip moves less than 10⁻⁹ px. The same joints without the declaration put it 68 px forward over the loop. With a slide of 0.04 leg lengths a frame, both feet are released as stepping and never re-plant without the declaration, and the supports alternate with it. Jumps are read as before.
A step releases the plant too: while the supporting ankle moves more than 0.03 leg lengths a frame, the other foot takes support if it is planted (not itself released or moving), and otherwise the hip follows the pose's travel. The released ankle may support again once it has moved less than 0.01 leg lengths a frame for 2 frames. On the dancer the plant held the supporting ankle through a real step (frames 176–181, 579 → 718 px) and threw the hip sideways.
Read from the lower ankle (#113, #119). On a sequence not declared "in-place", a foot whose ankle is at least 0.02 leg lengths higher in the picture than the other's never takes support: not when a plant starts, and not when a step releases the other foot. Before, a step handed support to the only foot free to take it, even a folded, lifted one, and the hip was then placed under a foot in the air. A supporting foot that the other ankle has been that much lower than for 0.1 s hands support to it when it may take it, as before; when the lower foot may not (it is released and not yet still, or moving), the supporting foot is released, and the hip follows the pose's travel until a foot may take support. On a frame after a jump or a step's release where the only foot free under #113 is the higher one (cause: "higher"), the lower foot supports tracked, not planted (#119): the hip puts it on #85's floor and follows its scaled ankle track (FEET tracked, feet.tracked) until it has been still for 2 frames (< 0.01), when the normal plant takes over at that position. Steps and jumps are read as before, and under "in-place" #92's reading stands. FEET support read from the lower ankle names every run of frames whose support differs from the reading before #113, the side before and after (none: no foot supports), and for each none why: released, the supporting foot let go because the lower ankle could not take support (released by a step, or moving), or the only foot free to take support was the higher one. The report's feet.lowered holds them as { frames, was, now, cause } (absent in place). On the tests' synthetic step under a lifted leg (the lifted ankle 0.2 leg lengths up, the grounded one stepping 0.05 leg lengths on frame 10 and still from 11), the lifted leg took support on frames 10–11 before; now frames 10–11 track the lower foot on the floor (feet.tracked) and the grounded one is planted from 12, when it is still again.
One floor (#85). A 2D rig has no depth, so its floor is one line, and a planted foot is on it. The picture disagrees: a step toward the camera puts the foot lower in the frame, and a camera residual puts a whole span higher, so a plant held where the picture puts each span leaves the floor stepping between spans. Every support span's supporting ankle is therefore moved in y onto one floor, the hip and everything above with it. A span's planted height is the world y (on the rig, posed by rig-c) its supporting ankle would be held at; the floor is the median of the spans' planted heights, or --floor <y> (world units, y up; only with --plant-feet, exit 2 without it). Each span's shift is floor − its planted height. A change of support still blends the hip over the 3 frames, now between two spans on the floor, and a planted foot does not leave the floor: on each blend frame the supporting leg is stretched along its chain about its root, by the factor --stance uses, k = (root y − floor y) / (root y − ankle y), held within [0.92, 1.08], its lower bound on the frame the supporting leg's own range as --stance's (#95; where that went below 0.92, a FEET bound from the leg's own range line, the report's feet.blend.range, and each blend run's lower). A frame that needs more is held at the bound, the rest left to the blend, and listed FEET past the bounds with the k needed and how far the foot was left from the floor. An ankle already within 0.01 units (the landing check's origin tolerance) is not stretched. The stretched scales are scale targets, so the landing check holds them; --stance then stretches the free leg, a different chain, to the supporting ankle. A jump follows the pose's travel from the hip it left, which is on the floor, carried linearly from take-off so the landing frame meets the floor and clamped so no frame inside the flight puts a foot below it (#103), and is held the same way over its first 3 planted frames, with the flight's carry held across the landing blend so it fades out with the blend weights rather than dropping on the first frame after landing (#147). --stance is unchanged: it stretches the free leg to the supporting ankle, which is on the floor. On the tests' synthetic stick, three spans 0, 6 and 1 px apart in the picture (0, 9 and 1.5 units on the rig) plant within 0.0001 units of one floor. With the shift forced to 0 they stay 9 units apart; without the stretch, the supporting ankle leaves the floor by 6.75 units on the first blend frame.
FEET lists the supports, the jumps and the steps (each with the side, its frames, who took support and where it re-planted; in place, the loop's closure instead), and with --plant-feet the floor: its world y and where it came from, each span's shift in units and in that side's rig leg at setup (thigh origin → ankle), and the shifts' range, then one line per run of blend frames whose supporting leg was stretched (blend held on the floor) and per run past the bounds. The report carries them under feet (supports, jumps, steps, changes, floor, shifts, blend, closure).
--stance <leg lengths> (only with --plant-feet, which it is refused without; off by default) plants both feet when the free one is near the floor (#60). On a frame with a supporting side (so outside every jump), the free ankle's height above the supporting one is read in the picture, in the free side's leg length — the leg length FEET uses. The tolerance is a distance (#114): within it, above the supporting ankle or below it, the frame is planted both: every bone of the free leg's chain has its scalex key multiplied by k = (root y − floor y) / (root y − ankle y) on the posed rig, the root being the chain's first bone's origin and the floor the supporting ankle, so the foot reaches the floor. A free ankle within the tolerance below the supporting one is planted too: its k is under 1, since a foot slightly under the floor is the same fault as one floating above it. A free ankle past the tolerance below the supporting one is not a stance but a support error (the foot read as supporting is the one in the air), so that frame is never planted both and keeps the pose's height; it is named on a STANCE past the tolerance below line with the side, the frames and the largest distance, and read "below" in the report's stance.runs. Before #114 the gap was signed, so any free ankle below the supporting one counted as within, and a lifted foot read as supporting had the grounded leg shrunk up to it. On the tests' synthetic stance, a free ankle 0.5 leg lengths below is read below and not planted; one 0.03 below is planted with k under 1. k is held within [0.92, 1.08], except that the lower bound comes from the leg's own range in the sequence (#95): per side, r = the leg's picture length hip → ankle on the frame ÷ its straight length (the 95th percentile of that length, as --leg-scale reads a segment's), and the frame's lower bound is min(0.92, r), so the rig leg may fold as far as the picture folds it on that frame and never less far than 0.92. The upper bound stays 1.08, since a picture leg does not read longer than straight. A sequence whose legs never fold below 0.92 keeps exactly [0.92, 1.08]. Where the range set a bound below 0.92 on a planted-both frame, one STANCE bound from the leg's own range line gives each side's lowest, as does the report's stance.range. A frame past its bound is listed past the bounds with the bound it had, held at the bound and not forced. Past the tolerance above, the foot is lifted and keeps the pose's height. The state is read with hysteresis and a shortest run (#87), on the distance: a planted foot leaves the stance when its distance from the supporting one goes past the tolerance, and one past it is planted again only under 0.8 of it (both figures are on the STANCE line). A change of state takes effect only when its condition holds for 0.1 s (by the frames' times). A shorter crossing with the other state on both sides of it keeps their state and is listed absorbed with its frames and the height that crossed (absorbed: true in the report's stance.runs, beside the run that took it in). One that runs to the end of a support span is taken, since nothing of the other state follows it. A threshold read frame by frame turned noise at the threshold into a pop: one frame at 0.0708 against 0.07 dropped the stretch for one frame and the foot jumped up and back. The stretched scales are the scale targets, so the landing check holds them. Measured with --stance 0.07 and HQ's flags (--smooth 7 --flip 90 --short 0.35 --bends clamp --leg-scale 0.8,1.05 --plant-feet --carry-idle): on HQ's dancer, frames 94–175 outside the jumps, the free ankle's height above the supporting one on the posed build went from median 18.3 and p90 25.4 units to 0.0 and 0.0. On a second sequence (Toca, 492 frames at 30 fps) the 29 frames labelled by eye as both feet flat, outside the jumps, read 0.0 and 0.0. Of the 6 labelled one foot lifted, 5 stay above the tolerance; the sixth (frame 360) was 0.042 leg lengths up in the picture, within the tolerance, and so planted; without --stance it stood 6.5 units up, also under the tolerance (22.6 units there). That is 83 %, under the 90 % the card asked.
The map needs one chain ending at each ankle whose last bone is aimed "x" (the ankle is that bone's tip), and a translate the legs hang from (LEGS_MAP). A range, or a set's turn or offsets, on a bone between the hip and an ankle is refused (LEGS_RANGE): the plant is derived before the range clamp, and a clamp after it would move the planted foot.

Both are checked on the gated build: every scale lands, each ankle lands where the keys put it, and the hip's origin where the plant put it. The figures were fixed in #47 before the dancer was measured; the defaults stay off until they pass.

The idle: what the retarget drops

A rig's motion spec usually has an idle — a blink, a breath, hair and hem sway — and the new animation keys none of it (#43, item 2). The run reads that animation (--idle <name> names another; naming one the spec lacks is IDLE_NAME) track by track against the new one, channel by channel — translate poses translatex and translatey, a group track poses each of its members, a slot's rgba its rgb and alpha:

IDLE-DROPPED — a track the new animation keys on none of its channels, and that turns no mapped bone (below): its target, property, keys and the idle's duration. The report lists them under idleDropped (target, members, property, keys, duration, carried, notCarried).
IDLE-KEPT-OUT — never carried (idleKeptOut), with the reason: a track the new animation keys itself, naming the track that keys it; or a rotate, scale or shear (any axis) of a bone above a mapped bone — between mapped bones, or above the first one (the map's chains and its translated bone) — naming the mapped bones below it: the retarget's keys hold every unmapped bone at its setup pose, so such a track would turn them off the angles the pose asks. A translate of such a bone is carried: it moves the bones under it and turns none. The exception is a translate of a bone above the translated bone (the hip), which is kept out because the retarget places that bone's origin. (Folding a carried parent's motion into the mapped bones' keys is a later card.) The idle's drawOrder, events, ik, transform, deform and sequence timelines are not tracks, and are listed here too.

--carry-idle (off by default: a blink keyed over a set's art may not suit it, and that is the author's call) loops every dropped track into the new animation: the idle's keys repeated every idle duration and cut at the new animation's end, with the value interpolated there.

A key's ease or curve describes the segment to the next key, so a loop is built key by key. Where the idle's last key is at its duration and its first at 0, the two are one key at each seam (the next cycle's first; if their values differ, the line says where the loop jumps, as a player looping the idle does). Where the last key is earlier, it is keyed stepped, so it holds to the cycle's end as it does in the idle. Where the first key is later than 0, each later cycle opens with a stepped setup key: 0 for a bone's translate, shear and rotate, 1 for its scale, the slot's setup attachment; for any other property such a track is not carried, and the line says why.
A raw curve states absolute times, so a repeated one is shifted with its key. The segment the cut falls in is split there: linear and stepped as they are, a Bézier (a raw curve, or a named easing turned into one as rigc does) by de Casteljau, so the part kept is the same cubic and the last key carries neither ease nor curve. rig-c draws a Bézier as Spine's polyline, so a split segment and the idle's whole one sit off the cubic, and off each other, by that alone (measured on the test's eased segment: the idle's own up to 2.6e-3, the carried part up to 9.3e-4).
A track with lag or stagger, or whose values are per-member maps or models, is not carried, and the line says so.

Every bone still has to land with the carried tracks in. A carried translate above the translated bone moves its origin off the pose's travel, and the run refuses with POSE_LANDS, naming the carried tracks above the bone that missed.

On HQ's dancer (--flip 90 --short 0.35 --bends clamp --carry-idle) the idle has 26 tracks. Kept out: the four arm rotates, which the dance keys, and a control bone's scalex above the chest, the neck's rotate and a head control's rotate, which sit above the mapped chest, head and arms — carried, they missed the chest, the head and the arms by 0.16°–1.37°. Carried: the other 19, the blink (an eyes group's scaley) and the breath (the chest control's translatey) among them. The gate is green, every bone lands (worst 0.00167°), and rig-c's poser shows the blink at every cycle (scaleY 0.12 at 2.37 s, 5.70 s, 9.08 s and 12.41 s, against the idle's at t mod 3.333 s within 8.9e-6) and the breath's offset (5.02 at 0.833, 4.167, 7.5, 10.833 s; 0 at 2.5, 5.833, 9.167, 12.5 s; 1.035 at the cut, 13 s).

--merge <file> merges a timed track fragment into the new animation, at absolute times from its start — blinks read off the video, for example:

{ "spec": "rig-pose-tracks/1", "tracks": [ { "group": "eyes", "property": "scaley", "keys": [ { "t": 2.3, "v": [1] }, { "t": 2.37, "v": [0.12] }, { "t": 2.45, "v": [1] } ] } ] }

Each track is a rigc-motion track. One posing a channel the new animation (or a carried idle track, or an earlier fragment track) already keys is refused by name (MERGE_CONFLICT): a fragment adds a track, it never replaces one. So is a key before 0 or past the animation's duration (MERGE_TIME), and a file of any other shape (MERGE_SPEC). What the track means — its target, property and values — is rigc's to judge: its gate checks every merged track like any other (RIG_GATE). MERGE lists each track merged; the report lists them under merged.

Few keys: `--reduce`

The retarget writes one key per pose frame on every mapped bone. That is faithful, but nobody can edit it by hand. --reduce <deg> (off by default) thins each mapped bone's rotate track to a few keys with curves, so that every bone's world aim stays within deg of the dense keys on every pose frame. --reduce-hip <units> does the same for the hip's translate, holding its world origin within units; without it, the hip stays dense.

Keys stay on the pose's own frame times. Between two kept keys, the span is linear if a straight line is within the tolerance. Otherwise it gets one raw Bézier curve, with its handles at the span's thirds and its values fitted by least squares. The error is measured the way rig-c's poser evaluates a curve (its 10-piece polyline). A span still over the tolerance gets a key at its worst frame, and both halves are fitted again.
Errors do not pile up down a chain. Bones are fitted top-down, and each bone is fitted against its dense key less the error its nearest mapped parent already carries. So a bone's world error is its own fit error.
Kept regardless: the first and last frame; the frames either side of a switch boundary, on the chains the switch changes; and the first and last frame of every FLIP spread, SHORT span and RANGE run, on that bone.
The landing check holds the dense targets with the tolerance added (0.01° + deg; 0.01 + units), and it is posed by rig-c as always.
On HQ's dancer (--smooth 7 --flip 90 --short 0.35 --bends clamp; 313 frames, 11 mapped bones): --reduce 1 --reduce-hip 1 takes the rotate keys from 3443 to 588 (5.86×), 296 of the spans curved, and the hip's translate keys from 313 to 96. The gate is green. rig-c's poser lands every bone within 0.9998° of the dense target (head) and the hip within 0.901 units. At 0.5° the keys drop 3.92×; at 2°, 8.38×.
It runs last, on the final keys (#94): after --plant-feet's plant, its one floor, the blend and --stance stretches and the in-place closure. Their keys are reduced like any other, and every landing check is run again on the reduced motion.
- A scaled bone's scalex track is thinned within 5e-5 of its setup length: half of the scale landing check's 1e-4, so a reduced curve cannot fail that check.
- A mapped bone under a scaled bone (keyed through its parent's posed matrix by --leg-scale) is solved again through its parent as reduced. Its rotate is then thinned with the error read as the world aim it lands on, and its scalex with the error read as the world length it lands on. Its world error is its own fit error, as down any other chain.
- Each ankle is held to the plant's tolerance plus the most the reduction may move it: the hip's tolerance (with --reduce-hip), plus the leg from the hip to the ankle (each bone's setup offset from its parent and the last bone's length) times the largest scale asked, times the angle tolerance in radians plus 5e-5. The ankle's LAND line prints that tolerance.

Exit codes: 0 written, 1 refused (every reason printed, nothing written), 2 bad invocation. rig-pose --help has every flag.

The pose: `rig-pose-sequence/1`

@rig-c/parts' keypoints.json (spine-parts-keypoints/1) extended to frames — the same vocabulary, one more level:

{ "spec": "rig-pose-sequence/1",
  "space": { "units": "picture-px", "origin": "top-left", "y": "down" },
  "width": 1080, "height": 1920,
  "source": "what produced the file",
  "frames": [
    { "t": 0, "people": [ { "id": "a", "joints": {
        "neck":       { "state": "observed", "at": [540, 410], "score": 0.93 },
        "r_shoulder": { "state": "observed", "at": [470, 430], "score": 0.91 },
        "r_wrist":    { "state": "missing" } } } ] } ] }

t is seconds and strictly increases; the animation's keys are at t − t₀.
Joints are body-18 (nose neck r_shoulder r_elbow r_wrist l_shoulder l_elbow l_wrist r_hip r_knee r_ankle l_hip l_knee l_ankle r_eye l_eye r_ear l_ear); r/l are the subject's sides. state is observed, occluded or missing, as in the keypoints file, or inferred (#63): a position the detector supplied but did not see. An inferred joint carries at (one without it is refused, POSE_JOINT_AT), and retarget keys it as it keys observed. An unlisted joint is missing.
Two differences from the keypoints file: the unit is picture-px, and at may lie outside the picture (a hand leaves the frame; the score says how much to believe it).
More than one person: --person <id>.
travel (optional, #92) is the clip's frame of reference. Its one value is "in-place": a gait filmed without the figure travelling, such as a treadmill or reel walk, or a loop. clean --loop writes it, and it may be written by hand on any sequence. Any other value is refused by name (POSE_TRAVEL). Without it the figure's travel in the picture is its travel. It changes only --plant-feet (below).
score is whatever the pose model gives. RTMPose's goes above 1, so --min-score is a cut on that model's scale, not a probability.

From COCO-17 — the layout most pose models and datasets use — rig-pose clean --coco in.json --out pose.json, which converts and cleans in one step. rig-pose convert --coco converts only, and its output still needs clean. retarget --coco is refused (POSE_NOT_CLEANED), naming the clean --coco command. The input is { "format": "coco17-sequence/1", "width", "height", "source", "frames": [{ "t", "keypoints" }] }, keypoints in COCO's order as 17 [x, y, score] or 51 numbers, or null for a frame with nobody. The mapping: COCO's left_*/right_* → body-18's l_*/r_* (both are the subject's sides), a score of 0 → missing, and body-18's neck, which COCO lacks, is the shoulders' midpoint scored the lower of the two (OpenPose's definition).

The camera (#63). A sequence may carry an optional camera, one { "s", "tx", "ty", "inliers" } per frame: the picture's scale and translation against frame 0's, so a point p₀ of frame 0's picture appears at s · p₀ + (tx, ty) on that frame. Its joints are then already normalised by it, p′ = (p − t) / s, and inliers is how many background features the fit kept. s must be > 0, tx and ty finite, inliers an integer ≥ 0, and there is one entry per frame; anything else is refused by name (POSE_CAMERA). The retarget does not apply it again: it prints a CAMERA line with the range of s and the three frames where it changes fastest. With --plant-feet and no camera, the CAMERA line says the floor may step with the picture, since a zoom or a pan in the video moves the floor with it. An optional cameraSource string says where the camera came from: "tracked", "hand-labelled, N keys" or "hand windows, floor-snapped: …". It is refused without a camera, or when it is not a non-empty string (POSE_CAMERA_SOURCE), and the CAMERA line starts with it (CAMERA tracked: …; a sequence with a camera and no cameraSource prints CAMERA (source not stated): …). tools/scan.py (not shipped) writes the normalised sequence. It reads the sequence before clean does (scan → clean → retarget): a sequence that already carries a cleaned record is refused by name (SCAN_AFTER_CLEAN), since its cleaning was done with the camera's steps still in it. The camera can come three ways, exactly one per run:

Tracked (--video), where the background has texture to follow. The camera is read off the video's background, and the sequence is written with "cameraSource": "tracked". A frame whose fit keeps too few background features is refused by name, and nothing is written.
A hand file (--camera-file), where the background cannot be tracked and the camera's values are known or judged by hand. The file is rig-pose-camera/1: { "spec": "rig-pose-camera/1", "keys": [ { "frame", "s", "tx", "ty" } ], "note"? }. Its frames are integers of the pose and strictly increase, s is > 0, and tx and ty are finite. Values are linear between keys and held before the first key and after the last. inliers is 0 on every frame, since no feature was fitted, and anything malformed is refused by name (CAMERA_FILE). The source is "hand-labelled, N keys".
Hand windows, snapped on the floor (--camera-windows a-b,c-d,…, #82), where the background cannot be tracked and the camera zooms in steps with a full-body figure on the floor. The windows are the frames of each zoom, by eye. Each step is measured on the floor, the one background feature every full-body clip has. An ankle is planted on a frame when it is still both ways, moving less than the retarget's stillness threshold (0.01 leg lengths a frame, in one picture leg: the clip median of hip → ankle) from the previous frame and to the next (through bun tools/supports.ts, beside the retarget's own jumps from supportsOf). Jump frames are not excluded: the detector reads them on the uncorrected picture, so they carry the very camera steps the windows remove, and an airborne ankle cannot hold still over three frames. A frame read inside a detector jump is named in the source (r 187 inside detector jump 187-193), not refused. For each window and each ankle, y before and y after are the ankle's median picture y on the 3 planted frames nearest the window on each side, searched within 15 frames and never past a neighbouring window, and its ratio about the picture centre c is (y_after − c_y) / (y_before − c_y). Where both ankles read, the step is the mean of the two, and their disagreement is always written. Where one reads, the window says which ankle alone. A window no ankle reads is refused by name (CAMERA_WINDOWS), and so are windows that are not a ≤ b inside the pose's frames, that overlap, or that leave no frame between them; nothing is written. s is 1 before the first window, multiplies by each step, is linear across a window and is held between windows, and t = c (1 − s), so no pan is read. The slow drift between windows (a push-in) is held, not fitted: that is open. The source is "hand windows, floor-snapped: a-b (r+l, l/r x.x %; r 83,84,85|97,100,104; l …), c-d (r alone; r …)" (no window for none): each window names the frames every median stood on, before and after the bar, and the CAMERA line prints it. The step moves with those frames: a 3-frame median moved ±2 % with the frames picked on a reference clip, which over three windows decided whether 37 or 107 of its 265 support frames left the floor band (#82). --plot draws both ankles' y with the windows shaded and the frames each window was read on marked, where a wrong window shows.

tools/scan_selftest.py is the synthetic check of all three.

The older name. rig-c-pose-frames/0, written by the extraction jobs this package's demos were made from, is the same file under another spec string. It has space.units "painting-px" and an fps key; each frame's t is the time read. It is read as it is.

The map: `rig-pose-map/1`

Which bones carry which joints. You write it for your rig:

{ "spec": "rig-pose-map/1",
  "chains": [
    { "name": "trunk", "bones": ["chest"], "joints": ["mid_hip", "neck"], "aim": ["neck"] },
    { "name": "arm_r", "bones": ["sleeve_r0", "sleeve_r1"], "joints": ["r_shoulder", "r_elbow", "r_wrist"] },
    { "name": "leg_l", "bones": ["leg_l0", "leg_l1"], "joints": ["l_hip", "l_knee", "l_ankle"] } ],
  "translate": { "bone": "hip", "joint": "mid_hip",
                 "scale": { "pose": ["mid_hip", "neck"], "rig": ["hip", "neck"] } } }

A chain's bone k carries the segment from joint k to joint k+1, so it lists one joint more than bones; each bone descends from the one before (a control bone between them is fine).
aim, one entry per bone: "x" (default) — the bone's own x axis is the segment; right for a limb link @rig-c/parts turned along its limb. A point — the direction from this bone's origin to that point; right for a trunk bone, whose own axis parts leaves at 0°. "a>b" — the direction from point a to point b; right for a head, turned by its eye line: { "bones": ["head"], "joints": ["r_eye", "l_eye"], "aim": ["eye_r>eye_l"] }, or for a pelvis, turned by its hip line: { "name": "pelvis", "bones": ["hip"], "joints": ["r_hip", "l_hip"], "aim": ["leg_r0>leg_l0"] }.
A point of the rig is a bone's origin (name), the end of its x axis (name:tip), or the midpoint of two of those ("a+b": "sleeve_r0+sleeve_l0" is the shoulders' midpoint). An aim's points must be the bone itself or descend from it.
reach: true, on a one-bone chain from the translated joint (a trunk from mid_hip): the bone is aimed from where its origin is posed — the hip moved and turned — at where the pose puts the segment's end relative to mid_hip, scaled. Without it, a tilted pelvis carries the chest's origin sideways and the trunk, parallel to the pose's, misses the shoulders. Its aim is one point, from the bone's own origin: { "name": "trunk", "bones": ["chest"], "joints": ["mid_hip", "mid_shoulder"], "aim": ["sleeve_r0+sleeve_l0"], "reach": true }. The chain must descend from the translated bone.
bend: every joint inside a chain is a hinge. "rest" (default) takes the side it may bend from the rig's rest pose and refuses a rest that is straight (under 0.5°); "ccw"/"cw" state it; "free" skips the check. ⚠️ Seen from the front, an elbow's side is not a joint limit. An elbow flexes toward the camera, and its picture falls on either side of the upper arm depending on how the arm is turned. On a real 13 s front-facing dance (313 frames, RTMPose), the right elbow sat on the "wrong" side of the rig's rest on 165 frames (median 23°), and the left on 200 (median 29°, up to 179° with the hands behind the head). Every one was a real elbow bend. The knees strayed at most 8.8° and 14.7°, which is pose noise around a straight leg. So for front-facing material, arms go "free". Legs on "rest" with --bends clamp hold only while the knees stay near straight. On a second dance with knee bends (Toca Toca, #27), that setting straightened the left knee for most of the dance: leg_l1 was 48.3° off at p90, against 0.0–0.3° with free legs. Leave legs "free" unless the dance keeps its knees straight.
mid_hip and mid_shoulder are the midpoints of the two hips and the two shoulders.
translate.scale is one length measured on both: two points on the rig, two points in the pose (its median over the frames where both are good). Pick a long segment that does not foreshorten much.
face (optional, #138): { "yaw"?: { "bone": "yaw", "range": [-15, 15] }, "pitch"?: { "bone": "pitch", "range": [-10, 15] } } (min ≤ 0 ≤ max, min < max). Reads per-frame yaw and pitch from nose against r_eye + l_eye (and r_ear + l_ear when present), compresses each axis into its declared range so peaks are scaled rather than clipped flat, and keys the named dial bone's rotate track (FACE).

A joint's range: `range`

A rig whose limbs are cut at the joints reads well only within some range of each joint's angle. A chain may state it per bone, in rotate-key degrees (the key, relative to the bone's setup rotation and to the turn of the mapped bone above it):

{ "name": "arm_r", "bones": ["sleeve_r0", "sleeve_r1"], "joints": ["r_shoulder", "r_elbow", "r_wrist"],
  "range": { "sleeve_r0": [-10, 5], "sleeve_r1": [-10, 10] } }

Outside every span of a set depicting that chain, a key past its range is clamped at the limit: the bone's target becomes the clamped world angle, and the bones below are keyed from it as usual, so they still point where the pose puts them. Bones are clamped parents first, and a reach below a clamped bone is aimed again from where the clamped bones put its origin. The landing check measures against the clamped targets. Every overflow is listed as RANGE demand. A key on its limit to within 0.001° (the writer's rounding of a key; the rest angles come from rig-c's float32 model) does not count. Inside a set's span the bone is not clamped: the set's art was painted for that pose. Its drawn angle, if the set gives one, is read against the range instead (ART-RANGE). Each name is refused if it is not a bone of the chain (MAP_RANGE), as is a range that is not [min, max] with min ≤ max.

Alternate drawings: `alternates`

Bone rotation cannot carry every pose: an arm behind the head, a forearm toward the camera, hands crossed in front of the body. The answer is a different drawing for that pose, swapped in. One key pose repaints every slot it moves (both arms, and the torso when the pose pulls it), so an alternate is a set of slots switched together. The map says which sets the rig has and the pose each was painted from; the run decides when each shows:

"alternates": [
  { "name": "raised_arms",
    "slots":   { "upperarm_r_raised": "raised/upperarm_r", "forearm_r_raised": "raised/forearm_r", "torso": "raised/torso" },
    "hides":   ["upperarm_r", "forearm_r"],
    "chains":  ["arm_r", "arm_l"],
    "pose":    { "r_shoulder": [400, 600], "r_elbow": [300, 450], "r_wrist": [450, 400],
                 "l_shoulder": [600, 600], "l_elbow": [700, 450], "l_wrist": [550, 400] },
    "drawn":   { "sleeve_r0": 120, "sleeve_r1": 30 },
    "offsets": { "sleeve_r1": [4, -2] } } ]

(The names, coordinates and angles above are illustrative.)

slots: what each slot of the set shows while it shows; each attachment must be in that slot under some skin of the rig, as rig-c built it. hides: the slots that are empty while it shows. A slot's place in the draw order is the rig's own, so a drawing that must sit in front of the torso is an attachment in a slot that is: a switch is attachment keys only, and no draw-order key is written.
chains: the map's chains whose pose the set depicts; it is selected by them. pose: the key skeleton the art was painted from, body-18 (or mid_hip/mid_shoulder) in picture px, y down, any scale and place — only its directions are read; every joint the listed chains use must be in it.
drawn (optional): each bone's world angle (Spine degrees) as painted. It changes no key: while a set shows, a bone is keyed to the pose as always, and the art turns with it. What it gives is a figure: the art turn, how far each bone is from the angle its art was painted at (artTurn, and ART-RANGE past the bone's range).
offsets (optional): a translate per bone, in its parent's axes, held while the set shows and 0 elsewhere (stepped keys), so the joints as painted are the pivots. A bone translate already keys (the hip) cannot take one.
place (optional): "by-hip" places a hand that meets the body by its offset from the hip (below).
Each name resolves against the built rig and the map, and a miss is refused by name, listing what exists: MAP_ALT_SLOT, MAP_ALT_ATTACHMENT, MAP_ALT_HIDES, MAP_ALT_CHAIN, MAP_ALT_JOINT, MAP_ALT_POSE, MAP_ALT_DRAWN, MAP_ALT_OFFSET (a bone that is not a bone of the set's chains) and MAP_ALT_PLACE.

When one shows. Per frame and per set, the distance is the largest difference, in degrees, between a segment of its chains as the retarget asked for it (after holds, the short hold, the flip guard and the hinge check; before any range clamp) and the same segment in its key skeleton. On each frame the nearest set within --switch-tolerance (default 25°) shows, ties going to the map's order; a run of one set shorter than --switch-min (default 0.25 s) is dropped and the base art shows there. Both figures were fixed before a run (#6). Sets compete only with those they share a slot (shown or hidden) or an offset bone with, so a drawing of each arm can show at once; two that compete never overlap.

Where a switch is keyed: --switch-at. By default (first) a boundary — the frame a set starts showing on, and the frame the base art or the next set comes back on — is where the selection above puts it: the first frame within the tolerance, often a frame where the arm is slowing into the pose, so the change of drawing can be seen on a nearly still arm. --switch-at peak moves each boundary to the frame within ±--switch-min/2 (±3 frames at 24 fps with the default 0.25 s) where the chains it changes turn fastest, so the cut lands in motion (#31). The speed on a frame is the largest |angular speed| in degrees per frame, a central difference of the target world angles (one-sided on the first and last frame), over every segment of the leaving set's and the arriving set's chains; it is read on the same targets as the distance, before the range clamp and any set's turn, which depend on where the spans are. The window is clipped so every span keeps --switch-min, a gap between two spans of one set keeps it too (the bridging holds), no boundary reaches its neighbour, and a set with frames shows only inside them. In it, a frame qualifies only if the set the move newly shows is within 2 × --switch-tolerance of its key skeleton on every frame it newly shows it on; the boundary goes to the fastest qualifying frame, and stays when none qualifies. A tie goes to the frame nearest the original, then the earlier. A boundary between two sets moves once.

The bound exists because speed alone rewarded showing art far from its pose. Unbounded, on HQ's dancer, the peak sat mid-swing: the largest distance over the frames a set showed rose from 25° to 135° (K2), 24° to 64° (K4), 21° to 59° (K1) and 21° to 38° (K3), the sets' turn held the arms hard against the pose (ART-RANGE runs 24 → 36), and the rig snapped to a hands-low drawing two frames before the arms came down. Bounded, every set stays within 50° (the largest is 43°, K4) and the median speed at the boundaries still rises from 9.7 to 14.2°/frame, but ART-RANGE runs still rise, 24 → 31 (K4 1 → 3, K1 6 → 8, K2 6 → 7, K3 2 → 4). That is why first is the default and peak an option (#31, round 2).

Either way the SWITCH line ends with moved on …, off …: each boundary's old and new frame (the same under first), the frames and seconds it moved, the speed on both and its window, and the report carries the same under moved.

What is written. An attachment track per slot whose picture changes: the set's attachment (or null for a hidden slot) where a span starts, and the slot's setup attachment, as the compiled model states it, where it ends. A direct switch from one set to another puts the first set's slots back to setup on the second's first frame, so an extra slot that is empty at setup goes empty. A switch onto frame i is keyed halfway between frames i−1 and i: a stepped key on the frame's own time fires a frame late wherever a player's accumulated time lands just below it (rig-c's docs/AUTHORING.md §4.5), and halfway both the pose's times and a renderer's i / fps see it on frame i. Before the first key a slot shows its setup attachment, so no key is written at 0 unless a span starts there. Offsets are translate keys at the same times, stepped. rig-c's gate checks these keys like any other.

TURN. The cutoff was measured on the three lines across the body before it was fixed (#6): on the 13 s front-facing dance their picture length never drops under 0.81 of its median and none reverses; on a second sequence the hip line falls to 0.17 at frames 465–491 (side-on) and the eye line to 0.22 at 26–58 (head in profile), and 3 frames reverse. 0.5 lies between. A line is reversed when it points more than 90° from the subject's right joint to the left (front-facing, it points at 0°). The lines are read after holds, held as the map's points are even where the map does not use them; a line whose points are good on no frame is named as not measured.

A stopgap window: frames. A set may state "frames": [[first, last], …]: it shows only inside those windows, and on them it is chosen over any set within tolerance that has no window. It is for two sets whose key skeletons the angles cannot tell apart. Measured on HQ's dancer, the hands clasped behind the back and the hands low in front differ by only 2–21°, and the wrist and hand-point scores do not separate them either; the cue that might is #33's depth state. It is recorded in the map as a stopgap, not a selection rule. A window restricts a set; it never forces one. Where the set does not show inside its window, the run says why in a FRAMES line: the stretch, and whether the pose there was past the tolerance (with the largest distance), or within it but the run was dropped.

How far a set's art may turn: turn. A drawing reads right only near the angle it was painted at. A set may state, per bone with a drawn angle, how far its art may turn from it, "turn": { "sleeve_r0": [-20, 20] }. While the set shows, the bone is held within it: its target becomes drawn + clamp(wrap(target − drawn), min, max), each bone on its own world angle, so a forearm under a held upper arm keeps its own target. The landing check measures against the held angle. The overflow the pose asked for is listed as ART-RANGE. A turn on a bone the set has no drawn angle for is refused (MAP_ALT_TURN).

A raised arm placed against the head: --raised-by-head. The pose's angles put a hand where it is against the dancer's own shoulder. On a rig whose shoulders sit nearer its head, or whose head is larger, the same angles put a hand raised beside the head on top of it. With --raised-by-head (off by default, #62) a set draws an arm raised when its drawn angle for the chain's first bone (the upper arm) is within 30° of straight up (a world angle of 60°–120°). While that set shows, the arm's wrist goes where the reference's is against the head:

The unit is the eye distance: on each side, its median over the span. That is the reference's |r_eye − l_eye|, and on the rig the distance between the two points the head chain's "a>b" aim names. The head centre is the eyes' midpoint on each frame. The head chain is the one carrying the eye line (r_eye → l_eye, either way round). The flag is refused without one, or with one not aimed "a>b" (RAISED_HEAD).
The target is rig head centre + (reference wrist − reference head centre) × (rig eye distance ÷ reference eye distance), the picture's y turned up, per frame.
The order of solving. The arm is solved last of the bones. By then the hip's travel, every reach, the head and every range clamp are final, and the arm moves neither the head nor its own shoulder. On each frame the rig's eye points and the shoulder are posed forward from those targets, as a reach's origin is: each rest offset is turned by the nearest mapped bone above.
Reach. The arm reaches the annulus about the shoulder from |l0 − l1| to l0 + l1. Here l0 runs from the upper arm's origin to the forearm's, and l1 from there to the wrist (the forearm's tip, or the point it aims at). A target outside it is replaced by the reach limit along the head-relative direction. That is the point inside the annulus, on the ray from the head centre through the target, nearest the target. Where the ray crosses the annulus twice, the nearer stretch wins, and a tie goes to the point nearer the head. A ray that never enters it gives the annulus point nearest the target. Every such frame is listed in REACH. A too-near frame is treated as a too-far one: the wrist is placed at the reach limit, and the distance the rule grades it at is the placed one. That is |placed − head centre|, not the reference's, which the arm cannot fold to.
The solve. Two bones on the rig's own lengths put the wrist on that point, the elbow bending to the side the pose's does. The set's turn is part of the solve. When it holds the upper arm, the forearm is re-aimed from the held elbow at the placed wrist rather than held on its own. ART-RANGE lists where the re-aimed forearm still cannot reach within turn, in two cases. One is a forearm past its own turn, held there. The other is a held upper arm whose re-aimed forearm leaves the wrist more than 0.01 eye distances short of its placed point; 0.01 is the float and rounding scale. That case is listed on the upper arm, with the shortfall in eye distances. The solved angles become the targets, which the keys and the landing check follow.
Frames outside those spans, and sets that draw no arm raised, are keyed as without the flag, byte for byte. A raised chain must have two bones, each aimed "x" or at one point, and the eye points must not turn with it (RAISED_CHAIN).

A hand that meets the body: "place": "by-hip". The same mismatch, pointed at the hip. On a clasped-hands set the pose's angles aim the forearms the right way, and the rig's arm lengths put the clasp lower or higher on the body than the video has it (#93). A set may ask for its hands to be placed against the hip instead, "place": "by-hip" (off by default; no flag). It is the solve above, with the head's measures swapped for the hip's:

The key skeleton carries both hips. A set with "by-hip" must have r_hip and l_hip in its pose: the map's author supplies them, where the hips were when the drawing was painted. A set without them is refused by name (MAP_ALT_PLACE, naming the missing hip); nothing is guessed from the chains' joints.
Which chains. A chain of the set is placed when it ends at a wrist (r_wrist, l_wrist) that the set's pose puts within one hip width of the mid-hip: |wrist − (r_hip + l_hip) / 2| ≤ |r_hip − l_hip|, read on the key skeleton. A set asking for "by-hip" with no such chain is refused, naming each wrist's distance in hip widths, and so is a place other than "by-hip", a pose whose hips meet, and a pose without a chain's wrist (MAP_ALT_PLACE; the missing wrist is also MAP_ALT_JOINT, as for any set).
The unit is the hip width: on each side, its median over the span. The reference's is |r_hip − l_hip|. The rig's is the distance between its two hip points, posed from the targets of the bones above as the eyes are: the two points a chain carrying the hip line (r_hip → l_hip, either way round) names with an "a>b" aim, a pelvis; else the origins of the first bones of the two chains starting at r_hip and at l_hip, the legs. A map with neither is refused (PLACE_HIP), and so is a span over which either median is 0 (PLACE_HIPS).
The target is the rig's mid-hip + (reference wrist − reference mid-hip) × (rig hip width ÷ reference hip width), the picture's y turned up, per frame. The order of solving, the reach, the two-bone solve inside the set's turn and ART-RANGE are those of --raised-by-head, with every eye distance a hip width (a held arm's shortfall too, its artRange entry marked anchor: "hip"), and the mid-hip where the head centre is.
Out of reach below the shoulder. The reach limit is taken on the ray from the mid-hip through the target. A hand at the hips sits on the far side of the mid-hip from the shoulder on most rigs, so that ray usually runs away from the arm and never enters its reach. The wrist then goes to the reach's point nearest the target, on the line from the shoulder (missed in REACH).
A chain both --raised-by-head and "by-hip" would place is refused (PLACE_BOTH). A placed chain must have two bones, each aimed "x" or at one point, and the hip points must not turn with it (PLACE_CHAIN).
Frames outside the set's spans, the chains it does not place, and every map without "place" are keyed as before, byte for byte. The report's byHip carries each span.

Compare: the rig's render against the video

rig-pose compare --reference ref.pose.json --render render.pose.json --map map.json \
  [--retarget-report report.json] [--reference-hands ref.wholebody.json --render-hands render.wholebody.json] \
  [--cloth-reference ref.cloth.json --cloth-render render.cloth.json --cloth-windows 3.9-7.4,7.6-11.1] \
  [--report compare.json] [--sheet compare.png]

Polishing a retarget used to mean rendering it, having a person look, and hearing what was wrong. Most of what a person names can be measured with the same pose detector on both sides (#38). compare reads two rig-pose-sequence/1 files on one time base: the reference (the source video's keypoints) and the render (the same detector run on rigc render frames of the retargeted motion). It lists the differences per body unit, worst first, each with the fix it suggests. The person is then needed only for taste. The detector stays outside rig-pose, as it does for retarget.

The figures and bars below are fixed. They were set in #38 before any render was measured, and they are grading lines for rig-c/exam. A change to one is a new comment on that issue first.

One time base. Both sequences have the same number of frames, and each frame's times differ by at most 1 ms. Render the motion at the reference's rate, frame for frame. Anything else is refused (COMPARE_FRAMES, COMPARE_TIMES, with the first frames that differ). The hands files and cloth tracks are held to the same rule against the reference (COMPARE_HANDS_FRAMES/_TIMES, COMPARE_CLOTH_FRAMES/_TIMES).

Positions are normalised per side. A position is divided by that side's own hip width: the median over its frames of |r_hip − l_hip|, in picture px. Directions are compared as Spine angles of the picture segment, |wrap(render − reference)|. Every joint with a position counts, whatever its score; only hand visibility reads scores. A frame where either side lacks a point the measure needs is not measured, and it is never over the bar.

A finding is a run of frames over a measure's bar lasting ≥ 0.15 s. A run is timed from the first frame's time to the end of the last frame, as END-ON is: 4 frames at 24 fps (0.167 s) are a finding, 3 (0.125 s) are not. Runs one frame apart merge, and the merged gap is not counted as over the bar. A finding's worst frame is the frame of its largest figure. Within a measure, findings are listed worst first.

unit	measure	figure, per frame	bar	fix class
head	eye-line tilt	\|direction difference\| of r_eye → l_eye	> 15°	not derivable
arms	direction, per mapped arm segment	\|direction difference\|	> 20°	paint inside a set's span, rig outside
arms	hand visibility, per hand	mean score of the hand's 21 points (or the wrist's score) on each side	reference ≥ 0.6 while render < 0.4	depth (#33)
body	torso length	\|mid_shoulder − mid_hip\| ÷ that side's median	the render's range < 0.5 × the reference's, over the whole run	motion
body	hip vertical travel	mid_hip's y in hip widths	the same rule as torso length	motion
legs	direction, per mapped leg segment	\|direction difference\|	> 20°	paint inside a set's span, rig outside
legs	foot plant, per ankle	the ankle's speed in hip widths per frame (from the frame before)	render > 0.05 while reference < 0.02	motion
cloth	lag, swing ratio, overshoot, per window	#34's three figures, on each side	outside #34's bands around the reference's figure	rig (physics)

Why these bars (#38):

20° for a segment's direction. HQ's rule for a set's art is drawn ± 20°: the art may turn up to 20° from the angle it was painted at and still read. A bigger difference is an arm that looks different.
0.6 and 0.4 for a hand. On the 13 s dance, the hands in front (S4) score 0.53–0.88 and the hands hidden behind the back (S1) score 0.41–0.57. 0.6 and 0.4 sit outside that overlap, one on each side.
15° for the head's tilt, 0.05 / 0.02 hip widths per frame for a sliding foot, and 0.5 × of the reference's range for a missing bounce were fixed with the others before the dancer was measured.

Units and segments. The map says which segments are compared. An arm segment joins two of one side's shoulder, elbow and wrist, and a leg segment two of one side's hip, knee and ankle. A segment's name is chain/bone. Other chains, such as the trunk, the pelvis or the head, are not compared as segments. The head is measured on its eye line whatever the map says, and the body on its torso and hip. A mapped arm or leg segment whose point has a position on no frame of one side is refused (COMPARE_JOINT_MISSING, naming the chain, the bone, the point and the side).

The fix class is given only when it can be derived:

An arm or leg direction. The finding's worst frame is checked against the spans in --retarget-report (its switches: each set's frames and the chains it depicts). Inside a span of a set that depicts the segment's chain, the class is paint: the drawing shown there is not this pose. Outside every such span it is rig: the art does not show the bone's direction. Without a report, a chain no set of the map depicts is rig. A chain some set depicts is not derivable, and the line names those sets.
Hand visibility is depth (#33): the hand is drawn behind something.
A foot slide or no bounce is motion.
Cloth is rig: the cloth is the rig's physics.
A head tilt has no class in #38, so the line says it is not derivable.

Hand visibility. rig-pose-sequence/1 carries body-18 only, so the hand points come from COCO-WholeBody files, one per side: --reference-hands and --render-hands. The shape is { "spec"?: "coco-wholebody-133", "fps"?, "width"?, "height"?, "source"?, "frames": [{ "t", "points": [[x, y, score] × 133] | null }] }. The subject's left hand is points 91–111 and the right hand 112–132. A frame's figure is the mean score of the hand's 21 points. A null frame (nobody found) is not measured. Without the files, the wrist's score is read instead: an absent or missing wrist reads 0, as COCO's v = 0 does. The summary line then says from wrist scores. One of the two files without the other is refused with exit 2, because the two sides must be measured the same way.

Body. Each figure is read over the whole run, so the unit gives at most one finding per measure, no bounce, spanning every frame. Its figure is the ratio of the ranges, render ÷ reference. Its worst frame, the one the sheet draws, is the reference's frame farthest from its median. The summary line always gives both ranges.

Cloth. Cloth is measured only when both cloth tracks are given, with --cloth-windows. Without them the unit reads not measured. A cloth track is HQ's rig-c-cloth-track/0: { "spec": "rig-c-cloth-track/0", "subject"?, "method"?, "source"?, "fps"?, "frames_count"?, "units"?, "hipWidth": { "median", "unit": "picture px" | "rig world units" }, "origin"?, "axes"?, "skirtSwing"?, "frames": [{ "i", "t", "hip_x", "hip_y", "hip_w", "tilt_deg", "hem_mid", "hem_l", "hem_r", "hem_l_y", "hem_r_y", "hem_band_cx", "skirt_cx", "skirt_top", "skirt_bottom" }] }, the frame keys all numbers. Every position is already divided by the subject's own median hip width, with x to the right. hip_x is the hips' midpoint and hem_mid the midpoint of the hem's leftmost and rightmost points in the lower half of the skirt (#34's hem); only those two and t are read. Any other spec, or a key outside that list, is refused (COMPARE_CLOTH). The three figures are #34's, as HQ's skirt criteria define them, ported from HQ's measuring script. On HQ's per-frame record of the dancer they reproduce its numbers exactly (reference A 0.958 / 2.634 / 0.061, B 0.696 / 1.409 / 0.000). Read through rig-c-cloth-track/0, whose values are rounded to 5 decimals, the reference's lag in A reads 0.959.

Both signals are smoothed with a Gaussian, σ = 1 frame over ± 4 frames, ends padded, before differentiating (numpy's gradient).
lag, in frames, hem after hip positive: the peak of the normalised cross-correlation of hip-x velocity with hem-x velocity over the window, searched over −6…+12 frames and refined by a parabola.
swing ratio: per hip stroke, the hem's excursion over the hip's. A stroke runs from one hip reversal (a zero crossing of hip velocity) to the next and moves at least the stroke threshold. A wiggle under the threshold is dropped as a pair. The hem is read round(lag) frames after each reversal. The figure is the median over the window's strokes.
overshoot: per stroke, how far the hem keeps travelling in the stroke's direction after the hips turn, within half of the next stroke (or 12 frames after the last), as a fraction of the hip stroke. The figure is the median.
The stroke threshold is 0.2 reference hip widths: one physical size on both sides (the criteria's Revision 1), expressed in the render's own hip widths. A render track in rig world units is carried by the retarget's scale, scale.factor in --retarget-report (rig units per reference px): 0.2 × reference hip width × factor ÷ render hip width. On the dancer that is 0.2 × 128.1 × 0.54152 ÷ 92.202 = 0.1505, the criteria's own figure. Without the report it is refused (COMPARE_CLOTH_SCALE). A render track in picture px is carried by the torso, as the retarget scales the rig: 0.2 × (reference hip width ÷ reference torso) ÷ (render hip width ÷ render torso), each torso the sequence's median mid_hip → mid_shoulder. A reference track not in picture px is refused (COMPARE_CLOTH_UNITS). The CLOTH lines print both thresholds and how the render's was carried.
The bands, per window, around the reference's own figure: lag ± 1.0 frame; swing ratio ± 25 %; overshoot ± 0.10, absolute, floored at 0. A render figure outside its band is a finding over the window's frames. A window with no stroke on one side is listed as not measured.
A window is t0-t1 in seconds, and windows are named A, B, … in the order given. A window's frames are round((t − t₀) × fps) up to the end, exclusive. A window of 12 frames or fewer is refused, since the lag search reaches 12.

Output. The first line gives the frames, the duration, both hip widths, and the report the spans came from. Then come one summary line per unit (HEAD, ARMS, BODY, LEGS, CLOTH) and one FIND line per finding. A summary line gives the unit's findings and, per measure, its findings, its worst figure with the subject and frame, the frames over the bar (in runs of any length), and the bar. A FIND line gives the unit, the measure, the subject and its points, the worst figure with both sides' readings and its frame, the frames and seconds, the frames over the bar, the bar, and the fix class with the reason. With cloth, a CLOTH line per window gives both sides' figures. The run ends with the number of findings.

--report <file> writes the same as JSON (rig-pose-compare/1):

inputs, frames, duration, hipWidth: { reference, render }, torso: { reference, render } (medians, px);
rule: { findingSeconds, findingGap, timeTolerance } and bars (every bar above, by name);
units: [{ unit, findings, notMeasured, measures: [{ measure, subjects, source, findings, overBar, notMeasured, bar, ranges? }] }];
findings: [{ unit, measure, subject, points, chain, frames: [first, last], seconds, over, worst: { frame, t, figure, reference, render }, figureUnit, bar, fix, fixWhy }];
cloth: null, or { windows: [{ name, t, frames, reference, render, band }], stroke: { reference, render, carried } }.

Numbers are rounded to 4 decimals, once, at the writer.

--sheet <png> draws the reference skeleton beside the render skeleton at each finding's worst frame, one row per finding, with the finding's own points marked in yellow. It never draws a video frame. A cloth finding is a figure over a window, with no single frame, so it has no row. When no finding has a frame, no sheet is written and the run says so.

The control is the reference compared with itself. It must report no finding in any unit, and every summary line says 0 findings. On HQ's dancer (313 frames, RTMPose, with its whole-body hand points and the retarget's report) it does.

Exit codes: 0 compared, whether it found anything or not, since a finding is not a failure; 1 refused, with every reason printed; 2 bad invocation.

Requirements

Bun ≥ 1.2 on PATH — rig-pose, like rig-c, runs its TypeScript sources directly. rig-c comes as a dependency. With @esotericsoftware/spine-core installed beside it, rig-c's gate runs the spine-core round trip; without it, rig-c's own gate, and the gate's report says which.

Why alternate drawings (0.2)

A flat painting turned by its bones matches each limb's direction, not its depth. On a real 13 s front-facing dance, retargeted onto a @rig-c/parts rig, five moments were found where it cannot follow. They were found by measuring the pose against the rig and checked against the video:

	t (s)	what the dancer does	why bone rotation cannot follow
S1	0.0–1.1	hands clasped behind the back	depth order: the forearms go behind the body
S2	1.0–2.25	a hand to the ear, then the forearm toward the camera	foreshortening (the upper arm at 0.27 of its length) and depth order
S3	3.7–7.5	hands behind the head	depth order; both arm meshes fold through the whole span
S4	7.7–10.9	hands crossed in front	which forearm is in front
S5	11.1–13.0	an arm thrust at the camera, then a V sign	foreshortening (0.63) and the shape of the hand; the arm mesh folds

The answer, since 0.2, is a second drawing of the arm for each of those poses, shown in place of the bent mesh while the pose lasts. rig-pose chooses when to show it, keys the switch in the motion, and lists every fold no drawing covers (#6, above).

What it does not do

Extract poses from video. The input is joint positions. How you get them is up to you: a pose model, a dataset, or by hand.
Turn the body. It works in 2D only. A segment pointing at the camera foreshortens, and the rig cannot: the direction is matched, the length is not. With --leg-scale, leg segments are the exception.
Read a moving camera. A reference whose camera zooms or pans moves the planted floor with it under --plant-feet (#52).
Front and behind, blinks, mouth shapes. The detector's scores and face points do not carry them (#33, #46).
Hands and fingers. There are no bones for them yet (rig-c/parts#171).

Licence

MIT, see LICENSE.

Name		Name	Last commit message	Last commit date
.github/workflows		.github/workflows
bin		bin
examples/wave		examples/wave
scripts		scripts
src		src
test		test
tools		tools
.gitignore		.gitignore
.release-please-manifest.json		.release-please-manifest.json
AGENTS.md		AGENTS.md
CHANGELOG.md		CHANGELOG.md
CLAUDE.md		CLAUDE.md
LICENSE		LICENSE
README.md		README.md
RELEASING.md		RELEASING.md
bun.lock		bun.lock
cli.ts		cli.ts
eslint.config.js		eslint.config.js
package.json		package.json
release-please-config.json		release-please-config.json
tsconfig.json		tsconfig.json

Navigation Menu

Repository navigation

About

Repository files navigation

@rig-c/pose

Try it

What the rig needs

Clean first: `rig-pose clean`

What it does

The legs: foreshortening and planted feet

The idle: what the retarget drops

Few keys: `--reduce`

The pose: `rig-pose-sequence/1`

The map: `rig-pose-map/1`

A joint's range: `range`

Alternate drawings: `alternates`

Compare: the rig's render against the video

Requirements

Why alternate drawings (0.2)

What it does not do

Licence

About

Releases

Packages

Contributors

Languages

Latest commit

History

Folders and files

Repository files navigation

@rig-c/pose

Try it

What the rig needs

Clean first: rig-pose clean

What it does

The legs: foreshortening and planted feet

The idle: what the retarget drops

Few keys: --reduce

The pose: rig-pose-sequence/1

The map: rig-pose-map/1

A joint's range: range

Alternate drawings: alternates

Compare: the rig's render against the video

Requirements

Why alternate drawings (0.2)

What it does not do

Licence

About

Topics

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages

Clean first: `rig-pose clean`

Few keys: `--reduce`

The pose: `rig-pose-sequence/1`

The map: `rig-pose-map/1`

A joint's range: `range`

Alternate drawings: `alternates`