likelyzhao / Sparse_rcnn_without_detectron2

Sparsercnn_without_detectron2

• Sparse R-CNN: End-to-End Object Detection with Learnable Proposals
• depend on https://github.com/PeizeSun/SparseR-CNN

requirements

• pycocotools
• fvcore
• PyTorch ≥ 1.5
• opencv
• torchvision==0.8.2

how to train

• mkdir -p datasets/coco
• ln -s /path_to_coco_dataset/annotations datasets/coco/annotations
• ln -s /path_to_coco_dataset/train2017 datasets/coco/train2017
• ln -s /path_to_coco_dataset/val2017 datasets/coco/val2017
• ./train_coco.sh

test

• ./test.sh

If it helps you, give me a star please