FFT-Based Selection and Optimization of Statistics for Robust Recognition of Severely Corrupted Images

2024-03-21

IEEE International Conference on Acoustics, Speech, and Signal Processing

AI Summary

To address the challenge of robust image classification for intelligent devices (e.g., robots) operating under severe real-world degradations—including noise, blur, and occlusion—this paper proposes FROST, a novel framework for test-time robustness. FROST is the first method to leverage high-frequency responses from the Fast Fourier Transform (FFT) to explicitly identify degradation types and dynamically select the most suitable normalization statistics (mean or variance) per network layer, enabling layer-adaptive batch normalization. Crucially, it requires no model retraining and is plug-and-play. Evaluated on the severely degraded ImageNet-C subset, FROST reduces the mean Corruption Error (mCE) from 40.9% to 25.7%, achieving a 37.1% relative improvement—the new state-of-the-art. This work establishes an efficient, general-purpose paradigm for robust visual inference at test time.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Intelligent Robots: Learning & Optimization for ROB

Application Category

• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

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.

Problem

• Enhancing model robustness for severely corrupted images
• Selecting optimal statistics via high-frequency feature analysis
• Improving classification accuracy under severe corruption conditions

Innovation

• Uses high-frequency features for corruption detection
• Selects layer-wise normalization statistics adaptively
• Optimizes statistics via FFT-based selection method