This repo is the official implementation of "DETRs with Collaborative Hybrid Assignments Training" by Zhuofan Zong, Guanglu Song, and Yu Liu.
In this paper, we present a novel collaborative hybrid assignments training scheme, namely Co-DETR, to learn more efficient and effective DETR-based detectors from versatile label assignment manners.
# Go to "toolbox/MMDetection" directory in root path
bash install_toolbox_mmdetection.sh
pip3 install -r requirements.txt
pip3 install urllib3==1.26.15
yum install -y mesa-libGL
Go to visit COCO official website, then select the COCO dataset you want to download.
Take coco2017 dataset as an example, specify /path/to/coco
to your COCO path in later training process, the unzipped dataset path structure sholud look like:
coco
├── annotations
│ ├── instances_train2017.json
│ ├── instances_val2017.json
│ └── ...
├── train2017
│ ├── 000000000009.jpg
│ ├── 000000000025.jpg
│ └── ...
├── val2017
│ ├── 000000000139.jpg
│ ├── 000000000285.jpg
│ └── ...
├── train2017.txt
├── val2017.txt
└── ...
# Make coco dataset path soft link to ./data/coco
mkdir data/
ln -s /path/to/coco ./data
# One GPU
export CUDA_VISIBLE_DEVICES=0
python3 train.py projects/configs/co_deformable_detr/co_deformable_detr_r50_1x_coco.py --work-dir path_to_exp --no-validate --auto-resume
# Eight GPUs
bash tools/dist_train.sh projects/configs/co_deformable_detr/co_deformable_detr_r50_1x_coco.py 8 path_to_exp --no-validate --auto-resume
# Evaluation
export CUDA_VISIBLE_DEVICES=0
PYTHONPATH=".:$PYTHONPATH" python3 tools/test.py projects/configs/co_deformable_detr/co_deformable_detr_r50_1x_coco.py path_to_exp/latest.pth --eval bbox
GPUs | FPS | Train Epochs | Box AP |
---|---|---|---|
BI-V100 x8 | 9.02 | 12 | 0.428 |
Dear OpenI User
Thank you for your continuous support to the Openl Qizhi Community AI Collaboration Platform. In order to protect your usage rights and ensure network security, we updated the Openl Qizhi Community AI Collaboration Platform Usage Agreement in January 2024. The updated agreement specifies that users are prohibited from using intranet penetration tools. After you click "Agree and continue", you can continue to use our services. Thank you for your cooperation and understanding.
For more agreement content, please refer to the《Openl Qizhi Community AI Collaboration Platform Usage Agreement》