Understanding deep features with computer-generated imagery
This work addresses the interpretability of deep learning features for computer vision researchers, but it is incremental as it builds on existing methods for analyzing CNNs.
The researchers tackled the problem of understanding how convolutional neural networks (CNNs) respond to scene factors like object style and lighting by controlling these factors using computer-generated imagery from 3D CAD models, and they quantified and visualized the importance of these factors across different CNNs and layers, showing that the analysis applies to natural images.
We introduce an approach for analyzing the variation of features generated by convolutional neural networks (CNNs) with respect to scene factors that occur in natural images. Such factors may include object style, 3D viewpoint, color, and scene lighting configuration. Our approach analyzes CNN feature responses corresponding to different scene factors by controlling for them via rendering using a large database of 3D CAD models. The rendered images are presented to a trained CNN and responses for different layers are studied with respect to the input scene factors. We perform a decomposition of the responses based on knowledge of the input scene factors and analyze the resulting components. In particular, we quantify their relative importance in the CNN responses and visualize them using principal component analysis. We show qualitative and quantitative results of our study on three CNNs trained on large image datasets: AlexNet, Places, and Oxford VGG. We observe important differences across the networks and CNN layers for different scene factors and object categories. Finally, we demonstrate that our analysis based on computer-generated imagery translates to the network representation of natural images.