FFT-based Selection and Optimization of Statistics for Robust Recognition of Severely Corrupted Images
This addresses the challenge of robust test-time performance for vision systems on smart devices like robotic agents, representing a strong specific gain rather than a broad paradigm shift.
The paper tackles the problem of improving model robustness on severely corrupted images by proposing FROST, a method that uses high-frequency features to detect corruption types and select layer-wise normalization statistics, achieving state-of-the-art results with up to 37.1% relative gain on ImageNet-C and improving the baseline mCE from 40.9% on severe corruptions.
Improving model robustness in case of corrupted images is among the key challenges to enable robust vision systems on smart devices, such as robotic agents. Particularly, robust test-time performance is imperative for most of the applications. This paper presents a novel approach to improve robustness of any classification model, especially on severely corrupted images. Our method (FROST) employs high-frequency features to detect input image corruption type, and select layer-wise feature normalization statistics. FROST provides the state-of-the-art results for different models and datasets, outperforming competitors on ImageNet-C by up to 37.1% relative gain, improving baseline of 40.9% mCE on severe corruptions.