Score: 0

Bidirectional Action Sequence Learning for Long-term Action Anticipation with Large Language Models

Published: August 1, 2025 | arXiv ID: 2508.00374v1

By: Yuji Sato, Yasunori Ishii, Takayoshi Yamashita

Potential Business Impact:

Predicts future actions by looking forward and backward.

Video-based long-term action anticipation is crucial for early risk detection in areas such as automated driving and robotics. Conventional approaches extract features from past actions using encoders and predict future events with decoders, which limits performance due to their unidirectional nature. These methods struggle to capture semantically distinct sub-actions within a scene. The proposed method, BiAnt, addresses this limitation by combining forward prediction with backward prediction using a large language model. Experimental results on Ego4D demonstrate that BiAnt improves performance in terms of edit distance compared to baseline methods.

Country of Origin
🇯🇵 Japan

Page Count
8 pages

Category
Computer Science:
CV and Pattern Recognition