Out-of-Distribution Detection with Semantic Mismatch under Masking

TL;DR

This paper introduces MoodCat, a novel OOD detection method that masks parts of an image, synthesizes the masked region conditioned on classification, and detects OODs by measuring semantic differences, outperforming existing methods.

Contribution

MoodCat is the first framework to leverage semantic mismatch via masking and conditional synthesis for effective OOD detection in image classifiers.

Findings

MoodCat significantly outperforms state-of-the-art OOD detection methods.

Semantic difference measurement effectively identifies out-of-distribution images.

Masking and conditional synthesis enhance the learning of in-distribution semantic features.

Abstract

This paper proposes a novel out-of-distribution (OOD) detection framework named MoodCat for image classifiers. MoodCat masks a random portion of the input image and uses a generative model to synthesize the masked image to a new image conditioned on the classification result. It then calculates the semantic difference between the original image and the synthesized one for OOD detection. Compared to existing solutions, MoodCat naturally learns the semantic information of the in-distribution data with the proposed mask and conditional synthesis strategy, which is critical to identifying OODs. Experimental results demonstrate that MoodCat outperforms state-of-the-art OOD detection solutions by a large margin.