Qinying-Liu / CASE

Accepted by ICCV2023, Revisiting Foreground and Background Separation in Weakly-supervised Temporal Action Localization: A Clustering-based Approach

Revisiting Foreground and Background Separation in Weakly-supervised Temporal Action Localization: A Clustering-based Approach

The code is assembled to OpenWTAL, which implements multiple WTAL methods in a unified codebase.

Revisiting Foreground and Background Separation in Weakly-supervised Temporal Action Localization: A Clustering-based Approach
Qinying Liu, Zilei Wang, Shenghai Rong, Junjie Li, Yixin Zhang
ICCV2023

[Paper]

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
   

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
   

   We provide our pre-trained checkpoints in checkpoints.zip

References

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