Probing Fine-Grained Action Understanding and Cross-View Generalization of Foundation Models

TL;DR

This paper systematically evaluates how different foundation models and design choices perform in fine-grained human activity recognition across varying camera perspectives, highlighting challenges and providing guidance for real-world applications.

Contribution

It offers the first comprehensive analysis of foundation models and temporal fusion strategies for cross-view fine-grained activity recognition.

Findings

Perspective changes significantly impact model accuracy.

Attention-based temporal fusion improves recognition across views.

Different backbone architectures vary in robustness to viewpoint changes.

Abstract

Foundation models (FMs) are large neural networks trained on broad datasets, excelling in downstream tasks with minimal fine-tuning. Human activity recognition in video has advanced with FMs, driven by competition among different architectures. However, high accuracies on standard benchmarks can draw an artificially rosy picture, as they often overlook real-world factors like changing camera perspectives. Popular benchmarks, mostly from YouTube or movies, offer diverse views but only coarse actions, which are insufficient for use-cases needing fine-grained, domain-specific actions. Domain-specific datasets (e.g., for industrial assembly) typically use data from limited static perspectives. This paper empirically evaluates how perspective changes affect different FMs in fine-grained human activity recognition. We compare multiple backbone architectures and design choices, including image- and video- based models, and various strategies for temporal information fusion, including commonly used score averaging and more novel attention-based temporal aggregation mechanisms. This is the first systematic study of different foundation models and specific design choices for human activity recognition from unknown views, conducted with the goal to provide guidance for backbone- and temporal- fusion scheme selection. Code and models will be made publicly available to the community.