How do vision transformers (ViTs) represent and process the world? This paper addresses this long-standing question through the first systematic analysis of 6.6K features across all layers, extracted via sparse autoencoders, and by introducing the residual replacement model, which replaces ViT computations with interpretable features in the residual stream. Our analysis reveals not only a feature evolution from low-level patterns to high-level semantics, but also how ViTs encode curves and spatial positions through specialized feature types. The residual replacement model scalably produces a faithful yet parsimonious circuit for human-scale interpretability by significantly simplifying the original computations. As a result, this framework enables intuitive understanding of ViT mechanisms. Finally, we demonstrate the utility of our framework in debiasing spurious correlations.

Most of previous interpolation methods cannot consider the sufficient edge information, and that result in unwanted artifacts. To remove or reduce the problem effectively, we propose a new image interpolation scheme based on Weighted Edge Pattern Analysis (WEPA). Proposed algorithm finds the accurate edge direction at current pixel by using the Spatial Direction Vector (SDV) which is result of WEPA of the neighboring pixels. After direction refinement, we interpolate the current pixel along obtained final SDV direction. Experimental results show that the proposed method exhibits a better performance than previous methods in both objective and subjective results with a variety of still images.