PeizeSun/SparseR-CNN

Health v2 · maintenance only

Adoption not part of the score

1343 stars · 191 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Sparse R-CNN is a PyTorch implementation (built on Detectron2) of the CVPR 2021 / PAMI 2023 paper 'End-to-End Object Detection with Learnable Proposals'. It provides a sparse, set-based object detection model that replaces dense candidate generation with a small set of learnable proposals, with pretrained models and training configs included.

Use cases

• run end-to-end object detection with learnable proposals
• train a sparse detection model on custom datasets
• reproduce CVPR 2021 Sparse R-CNN benchmark results
• compare sparse detectors against DETR and Faster R-CNN
• download pretrained COCO detection checkpoints
• research learnable proposal-based detection architectures

When to choose

• you need a sparse, end-to-end object detector without dense candidate generation
• you want to reproduce or extend published Sparse R-CNN results
• you already use Detectron2 and want a set-based detection head
• you need pretrained COCO models with reported AP and inference speeds

When to avoid

• you need a maintained production detector - the repo has seen little activity since 2023
• you want plug-and-play detection without GPU training infrastructure
• you need detection on mobile or edge devices
• you require non-PyTorch or non-Detectron2 ecosystems

Facets

library · maturity maintenance

machine-learning deep-learning computer-vision image-processing computer-vision deep-learning machine-learning artificial-intelligence python object-detection sparse-r-cnn cvpr-2021 detectron2 learnable-proposals computer-vision-research pytorch gpu linux docker

Member repositories

For agents

markdown · JSON · MCP: product_card(name="PeizeSun/SparseR-CNN")