This is the implementation for Hierarchical Graph Convolutional Networks for Action Quality Assessment, which has been accepted by IEEE TCSVT. The framework overview is as follows:
- We provide the dependency file of our experimental environment, one can install all dependencies by creating a new anaconda virtual environment and running:
pip install -r requirements.txt- torch_videovision
pip install git https://github.com/hassony2/torch_videovision- MTL-AQA: Download the dataset from the repository. The data structure is:
$DATASET_ROOT
├── MTL-AQA/
├── new
├── new_total_frames_256s
├── 01
...
└── 09
├── info
├── final_annotations_dict_with_dive_number
├── test_split_0.pkl
└── train_split_0.pkl
└── model_rgb.pth- AQA-7: Download AQA-7 Dataset using the following commands:
mkdir AQA-Seven & cd AQA-Seven
wget http://rtis.oit.unlv.edu/datasets/AQA-7.zip
unzip AQA-7.zipThe data structure is:
$DATASET_ROOT
├── Seven/
├── diving-out
├── 001
├── img_00001.jpg
...
...
└── 370
├── gym_vault-out
├── 001
├── img_00001.jpg
...
...
└── Split_4
├── split_4_test_list.mat
└── split_4_train_list.mat- JIGSAWS: Download the dataset from the website.
The command helper is:
usage: main.py [-h] [--benchmark {MTL,Seven,JIGSAWS}] [--exp_name EXP_NAME]
[--fix_bn FIX_BN] [--resume RESUME] [--ckpts CKPTS]
[--class_idx {1,2,3,4,5,6}] [--use_gpu USE_GPU]
[--device DEVICE] [--seed SEED] [--model MODEL] [--phase PHASE]
[--config CONFIG] [--weight_decay WEIGHT_DECAY]
[--base_lr BASE_LR] [--fold FOLD] [--bs_train BS_TRAIN]
[--bs_test BS_TEST] [--num_groups NUM_GROUPS]One can use the following command to train the model from scratch:
python main.py --benchmark MTL/Seven --phase train \
--model models.hgcn --device 01 --exp_name default \
--num_groups 1/2/4/8 --ckpts weights/MTL.pthWe provide the pretrained model at Google drive. Please download it into the ./weights folder. One can use the following command to run the example code:
python main.py --benchmark MTL --phase test \
--model models.hgcn --device 01 \
--num_groups 8 --ckpts weights/MTL-group8.pthIf you find our work useful in your research, please consider citing:
@article{zhou23hierarchical,
author={Zhou, Kanglei and Ma, Yue and Shum, Hubert P. H. and Liang, Xiaohui},
journal={IEEE Transactions on Circuits and Systems for Video Technology},
title={Hierarchical Graph Convolutional Networks for Action Quality Assessment},
year={2023},
doi={10.1109/TCSVT.2023.3281413},
issn={1051-8215},
publisher={IEEE},
}For any questions, feel free to contact me via [email protected].