duclvq/improved_ddnet

improved_DDNet

installation

# Clone the repository
git clone https://github.com/duclvQ/improved_DDNet.git
# Navigate to the project directory
cd improved_DDNet
pip -r install requirements.txt

download train/test.pickle from here to folder "data", and the "pose_new_v2" from here:

location.

./data/COBOT/train.pickle
./data/COBOT/test.pickle
./pose_new_v2

training

python train.py  --run_name {run name} #  find the list of "run_name" in "models/DDNet_Original.py" 

evaluation

find pretrained weights at this 'link'.

online test

Visit 'online.py' then set the value of 'exp' corresponding to run_name in the previous step, this action will return a folder of numpy files for all videos in test list.

After the prediction step, the frame-wise Acc is measured in 'online_local.ipynb', remember to read and set carefully every variable in these 2 files when you run.

draw_cfm.ipynb and visualize.ipynb also contain helpful functions.