Out-of-Distribution Detection with Semantic Mismatch under Masking
This addresses the problem of identifying out-of-distribution images for image classifiers, which is crucial for safety in real-world applications, but appears incremental as it builds on existing OOD detection methods with a novel masking and synthesis approach.
The paper tackles out-of-distribution detection for image classifiers by proposing MoodCat, which masks parts of input images and uses a generative model to synthesize new images based on classification results, then detects OODs by calculating semantic differences; it reports outperforming state-of-the-art solutions by a large margin.
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.