README

1. download code

git clone https://gitee.com/one2l/sparsercnns.git

2. Install cupy

pip install cupy-cuda113

3. Install detectron2

cd sparsercnn
python -m pip install -e detectron2

4. Make softlink for crowdhuman datasset

for linux environment

ln -s [dataset path] [project datasets path]

ln -s ~/autodl-tmp/COCOCrowdHuman/annotations ~/autodl-tmp/srcnn/exp/datasets/crowdhuman/annotations
ln -s ~/autodl-tmp/COCOCrowdHuman/CrowdHuman_train/Images ~/autodl-tmp/srcnn/exp/datasets/crowdhuman/CrowdHuman_train
ln -s ~/autodl-tmp/COCOCrowdHuman/CrowdHuman_val/Images ~/autodl-tmp/srcnn/exp/datasets/crowdhuman/CrowdHuman_val

for win environment

mklink /d [project datasets path] [dataset path]

mklink /d F:\workspace\sparsercnn\projects\datasets\crowdhuman\annotations D:\Datasets\COCOCrowdHuman\annotations 
mklink /d F:\workspace\sparsercnn\projects\datasets\crowdhuman\CrowdHuman_train D:\Datasets\COCOCrowdHuman\CrowdHuman_train\Images
mklink /d F:\workspace\sparsercnn\projects\datasets\crowdhuman\CrowdHuman_val D:\Datasets\COCOCrowdHuman\CrowdHuman_val\Images 

5.Train

# train baseline
python train_baseline.py --num-gpus 1 --config-file configs/sparsercnn.crowdhuman.res50.500pro.68e.yaml OUTPUT_DIR output/output_baseline

6. Test

python train_baseline.py --num-gpus 1 --config-file configs/sparsercnn.crowdhuman.res50.500pro.50e.yaml --eval-only MODEL.WEIGHTS output/output_vbox_crowdhuman/model_points.pth

7. Eval

python  crowdhuman-evl/crowdhuman_eval.py --result output/output_vbox_crowdhuman/inference/coco_instances_results.json --gt datasets/crowdhuman/annotations/val.json