Two-path Target-aware Contrastive Regression for Action Quality Assessment

This repository contains the PyTorch implementation for Two-path Target-aware Contrastive Regression.

Code for Two-path Target-aware Contrastive Regression (T²CR)

Requirement

• Python 3.7.9
• Pytorch 1.7.1
• torchvision 0.8.2
• timm 0.3.4
• torch_videovision

  pip install git+https://github.com/hassony2/torch_videovision

Data Preperation

• The FineDiving dataset [Xu et al.]
• The MTL-AQA dataset [Parmar and Morris]
• Download AQA-7 Dataset:

  mkdir AQA-Seven & cd AQA-Seven
  wget http://rtis.oit.unlv.edu/datasets/AQA-7.zip
  unzip AQA-7.zip

Pretrain Model

The Kinetics pretrained I3D downloaded from the reposity kinetics_i3d_pytorch

model_rgb.pth

Experimental Setting

MTL_T2CR.yaml
FineDiving_T2CR.yaml
Seven_T2CR.yaml

Training and Evaluation

# train a model
bash ./scripts/train.sh 0,1 MTL T²CR

# resume the training process
bash ./scripts/train.sh 0,1 MTL T²CR --resume

# test a trained model
bash ./scripts/test.sh 0,1 MTL T²CR ./experiments/T2CR/MTL/T²CR/last.pth
# last.pth is obtained by train.sh and saved at "experiments/T2CR/MTL/T²CR/"

More descriptions coming soon!