Understanding Transformer-based Vision Models through Inversion

TL;DR

This paper introduces an efficient feature inversion method to interpret transformer-based vision models, revealing how they encode image features, shape, and details, thereby deepening understanding of their internal representations.

Contribution

The study presents a novel, modular feature inversion technique applied to large-scale transformer vision models, enabling qualitative and quantitative analysis of their internal representations.

Findings

Models encode contextual shape and image details.

Layer correlations reveal internal structure.

Robustness against color perturbations is analyzed.

Abstract

Understanding the mechanisms underlying deep neural networks remains a fundamental challenge in machine learning and computer vision. One promising, yet only preliminarily explored approach, is feature inversion, which attempts to reconstruct images from intermediate representations using trained inverse neural networks. In this study, we revisit feature inversion, introducing a novel, modular variation that enables significantly more efficient application of the technique. We demonstrate how our method can be systematically applied to the large-scale transformer-based vision models, Detection Transformer and Vision Transformer, and how reconstructed images can be qualitatively interpreted in a meaningful way. We further quantitatively evaluate our method, thereby uncovering underlying mechanisms of representing image features that emerge in the two transformer architectures. Our analysis reveals key insights into how these models encode contextual shape and image details, how their layers correlate, and their robustness against color perturbations. These findings contribute to a deeper understanding of transformer-based vision models and their internal representations. The code for reproducing our experiments is available at github.com/wiskott-lab/inverse-tvm.