NMS-Loss: Learning with Non-Maximum Suppression for Crowded Pedestrian Detection
NMS-Loss test on Citypersons和Caltech：
| dataset | Config | MR |
|---|---|---|
| Citypersons | cityperons.py | 10.08% |
| Caltech(Ori) | caltech.py | 5.92% |
Prerequisites:
- Linux (Windows is not officially supported)
- Python 3.5+
- PyTorch 1.1
- CUDA 9.0 or higher
- NCCL 2
- GCC 4.9 or higher
- mmcv==0.2.16
a. Create a conda virtual environment and activate it.
conda create -n nms-loss python=3.7 -y
conda activate nms-loss
b. Install PyTorch, torchvision and mmcv
conda install pytorch=1.1.0 torchvision
pip install "git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI"
pip install mmcv==0.2.16
c. Clone
git clone http://git.code.oa.com/zekunluo/nms-loss.git
cd nms-loss
d. Check GCC, if GCC < 4.9:
conda install -c psic4 gcc5
e. install
./compile.sh
pip install -v -e . # or "python setup.py develop"
Dowdload weights from https://drive.google.com/drive/folders/1MwdnknqX6I3lNIbMVJQOyVxGK1lw-dEX?usp=sharing.
Citypersons:
./tools/dist_test.sh configs/cityperons.py work_dirs/citypersons.pth 8 --out results/citypersons.pkl --eval bbox
python3 tools/eval_script/eval_demo.py
Caltech:
./tools/dist_test.sh configs/caltech.py work_dirs/caltech.pth 8 --out results/caltech.pkl --eval bbox
python3 tools/caltech_pkl2txt.py