Blockwise Temporal-Spatial Pathway Network

TL;DR

This paper introduces BTSNet, a 3D-CNN model for video action recognition that adaptively adjusts spatial and temporal receptive fields, improving performance and interpretability across multiple datasets.

Contribution

The paper presents a novel blockwise temporal-spatial pathway network that adaptively selects receptive fields and fuses attention-based features for enhanced action recognition.

Findings

Achieved strong generalization on UCF-101, HMDB-51, SVW, and Epic-Kitchen datasets.

Provided interpretable visualizations of spatiotemporal attention.

Demonstrated improved representation for 3D convolutional blocks.

Abstract

Algorithms for video action recognition should consider not only spatial information but also temporal relations, which remains challenging. We propose a 3D-CNN-based action recognition model, called the blockwise temporal-spatial path-way network (BTSNet), which can adjust the temporal and spatial receptive fields by multiple pathways. We designed a novel model inspired by an adaptive kernel selection-based model, which is an architecture for effective feature encoding that adaptively chooses spatial receptive fields for image recognition. Expanding this approach to the temporal domain, our model extracts temporal and channel-wise attention and fuses information on various candidate operations. For evaluation, we tested our proposed model on UCF-101, HMDB-51, SVW, and Epic-Kitchen datasets and showed that it generalized well without pretraining. BTSNet also provides interpretable visualization based on spatiotemporal channel-wise attention. We confirm that the blockwise temporal-spatial pathway supports a better representation for 3D convolutional blocks based on this visualization.