Feature Visualization in 3D Convolutional Neural Networks

TL;DR

This paper introduces a new visualization method for 3D convolutional neural network kernels that separates texture and motion preferences, enhancing interpretability of complex 3D features.

Contribution

We propose a novel data-driven, two-stage optimization approach to disentangle texture and motion components in 3D CNN kernel visualizations, improving interpretability.

Findings

Visualizations reveal motion preferences of 3D kernels

Method provides clearer insights into 3D convolutional features

Effective across various pre-trained models

Abstract

Understanding the computations of convolutional neural networks requires effective visualization of their kernels. While maximal activation methods have proven successful in highlighting the preferred features of 2D convolutional kernels, directly applying these techniques to 3D convolutions often leads to uninterpretable results due to the higher dimensionality and complexity of 3D features. To address this challenge, we propose a novel visualization approach for 3D convolutional kernels that disentangles their texture and motion preferences. Our method begins with a data-driven decomposition of the optimal input that maximally activates a given kernel. We then introduce a two-stage optimization strategy to extract distinct texture and motion components from this input. Applying our approach to visualize kernels at various depths of several pre-trained models, we find that the resulting visualizations--particularly those capturing motion--clearly reveal the preferred dynamic patterns encoded by 3D kernels. These results demonstrate the effectiveness of our method in providing interpretable insights into 3D convolutional operations. Code is available at https://github.com/YatangLiLab/3DKernelVisualizer.