temporal_action_segmentation

Temporal Action Segmentation

This source code is for the temporal action segmentation (TAS) evaluation of twelvelabs embeddings

Embedding Preparation

• Download the embeddings via following links.
• Supported bechmarks
  • 50Salads (link)
  • GTEA (link)
  • Breakfast (link)

Run

After preparing the embeddings, you can run knn by following command.

python main.py --embeddings_dir [embeddings_dir] --dataset_name [dataset_name]

• embeddings_dir should be the parent directory of 'train' folder.
• dataset_name should be one from the following list.
  • 50Salads (50 Salads Dataset)
  • GTEA (GTEA, Georgia Tech Egocentric Activity Datasets)
  • Breakfast (Breakfast Action Dataset)

Reference

This TAS evaluation source code is based on Video Mamba Suit