Video Action Transformer Network: Implementation and Experiments

Explore the Action Transformer Network in this 21-minute video from the University of Central Florida. Learn about the proposed approach for action recognition, including the trunk region, head region, and multi-head, multi-layer transformer head. Dive into implementation details such as the Region Proposal Network and Attention Mechanism. Examine experiments conducted on various datasets, including the number of training examples per class, box area analysis, and boxes in a clip. Gain insights from qualitative results, including incorrect predictions, and understand the conclusions drawn from this research on action recognition in video content.

Syllabus

Intro
Outline
Objective
Proposed Approach
Trunk Region
Head Region Multi-head, multi-layer Tx Head
Action Transformer Network
Implementation Details
Region Proposal Network
Attention Mechanism
Experiments - Dataset Description
Experiments - No. of Training example: class
Experiments - Box Area
Experiments - Boxes in a clip
Qualitative Results - Incorrect Predicti
Conclusion