UniWTAL

Introduction

A unified and simple codebase for weakly-supervised temporal action localization, which currently contains the implementation of ASL(CVPR21), AICL(AAAI23), CASE(ICCV23)

Data Preparation

1. Download the features of THUMOS14 from dataset.zip.
2. Place the features inside the ./data folder.

Train and Evaluate

1. Train the CASE model by run

   python main_case.py --exp_name CASE
   

   Train the ASL model by run

   python main_asl.py --exp_name ASL
   

   Train the AICL model by run

   python main_aicl.py --exp_name AICL
   

2. The pre-trained model will be saved in the ./outputs folder. You can evaluate the model by running the command below.

   python main_case.py --exp_name CASE --inference_only
   

   python main_asl.py --exp_name ASL --inference_only
   

   python main_aicl.py --exp_name AICL --inference_only
   

   We provide our pre-trained checkpoints in checkpoints.zip

Todo

1. Code for ActivityNet
2. Code for more methods, e.g., C3BN, BAS-Net

References

• https://github.com/Pilhyeon/BaSNet-pytorch
• https://github.com/layer6ai-labs/ASL