This is a repository for PANDA(Gigapixel-Level Human-Centric Video Dataset) competition
We assume the root path is $SOTS, e.g., /home/chaoliang/SOTS
If you have set up environment according to CSTrack_PANDA tutorial, please skip here.
conda create -n CSTrack python=3.8
source activate CSTrack
conda install pytorch=1.7.0 torchvision cudatoolkit=11.0 -c pytorch
conda install -c anaconda mpi4py==3.0.3 --yes
cd $SOTS/lib/tutorial/CSTrack_panda/
pip install -r requirements.txt
- Download the yolov5 model[Baidu NetDisk(q8j9)] trained on tianchi-PANDA preliminary competition datasets to
$SOTS/yolov5_panda/weights
- Download the PANDA datasets[Baidu NetDisk(ecxm)] to
$SOTS/tcdata
, e.g.,$SOTS/tcdata/panda_round2_train_20210331_part10
cd $SOTS/yolov5_panda
mpirun -np 1 python detect_mpi.py --iou_thres 0.5 \
--conf_thres 0.4 \
--weights weights/yolov5_panda.pt
- Download the yolov5 pretrained model which pretrains on COCO dataset to
$SOTS/yolov5_panda/weights
- Download tianchi-PANDA preliminary competition datasets to
$SOTS/tcdata
Note: yolov5 training datasets are mere detection datasets of the preliminary competition
cd $SOTS/lib/utils/panda
python label_clean.py
mpirun -np 2 python split_det.py
cd $SOTS/yolov5_panda
python train.py --device 0,1 --batch-size 48