lvis

Real-time and accurate open-vocabulary end-to-end object detection

Continuation of an abandoned project fast-coco-eval

This repository contains the implementation for the paper "Revisiting Few Shot Object Detection with Vision-Language Models"

D2Det using mmdetection v2.1, supporting Objects365 and LVIS

DiverGen (CVPR 2024) & BSGAL (ICML 2024)

[ICCV 2021] MosaicOS: A Simple and Effective Use of Object-Centric Images for Long-Tailed Object Detection

[ECCV2022] Gumbel Optimised Loss for Long Tailed Instance Segmentation.

Perception evaluation toolkit in Rust with Python bindings. Drop-in pycocotools replacement, bit-identical and up to 113× faster, plus panoptic and semantic segmentation, TIDE error analysis, confusion matrices, calibration, model comparison, and a dataset browser.

[TIP 2023] Inverse Image Frequency for long-tailed image recognition.

Grid laser-shot-based LVIS L2 data from NASA into raster data

OVDeploy-Bench: EpisodicAP v2 + OOV-FP for deployment-style open-vocabulary detection

DETR + LoRA localises trucks, a CLIP-feature head classifies them: two-stage AP 0.514 vs 0.397 for DETR alone on 4 LVIS truck classes.

A YOLO instance segmentation model in practice. Identifying and outlining common objects in images.