Spatial-temporal Concept based Explanation of 3D ConvNets

TL;DR

This paper introduces a novel framework for interpreting 3D ConvNets in video recognition by using supervoxels and voxel importance scores, enabling understanding of spatial-temporal concepts influencing decisions.

Contribution

It proposes the 3D ACE framework that leverages high-level supervoxels and voxel scoring to interpret 3D ConvNets, addressing the gap in explainability for video data.

Findings

Discover spatial-temporal concepts of varying importance levels

Explore influence of concepts on action classification

Framework effectively interprets 3D ConvNet decisions

Abstract

Recent studies have achieved outstanding success in explaining 2D image recognition ConvNets. On the other hand, due to the computation cost and complexity of video data, the explanation of 3D video recognition ConvNets is relatively less studied. In this paper, we present a 3D ACE (Automatic Concept-based Explanation) framework for interpreting 3D ConvNets. In our approach: (1) videos are represented using high-level supervoxels, which is straightforward for human to understand; and (2) the interpreting framework estimates a score for each voxel, which reflects its importance in the decision procedure. Experiments show that our method can discover spatial-temporal concepts of different importance-levels, and thus can explore the influence of the concepts on a target task, such as action classification, in-depth. The codes are publicly available.