FAST: Foreground-aware Diffusion with Accelerated Sampling Trajectory for Segmentation-oriented Anomaly Synthesis

2025-09-24

AI Summary

Industrial anomaly segmentation suffers from scarce pixel-level annotations, highly diverse anomaly morphologies, and prohibitively high annotation costs. Existing synthetic anomaly generation methods struggle to balance sampling efficiency and generation fidelity, overlook statistical disparities between anomalous and background regions, and lack structural controllability. To address these challenges, we propose FADiff—a foreground-aware diffusion framework. It introduces a Foreground-Aware Reconstruction Module (FARM) to explicitly model anomalies and backgrounds with distinct priors; incorporates training-free Anomaly-Guided Iterative Acceleration Sampling (AIAS), enabling high-fidelity synthesis in ≤10 steps; and integrates mask-guided noise modulation with coarse-to-fine feature aggregation for fine-grained, adaptive control over anomaly regions. Extensive experiments demonstrate that FADiff significantly outperforms state-of-the-art methods across multiple benchmarks: synthesized anomalies better approximate real-world distributions, and downstream segmentation accuracy improves markedly. To our knowledge, FADiff is the first diffusion-based approach achieving efficient, controllable, and structurally specific industrial anomaly synthesis.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Machine Learning: Deep Generative Models & Autoencoders

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Web Mining and Content Analysis: Web data generation and simulation

Abstract

Industrial anomaly segmentation relies heavily on pixel-level annotations, yet real-world anomalies are often scarce, diverse, and costly to label. Segmentation-oriented industrial anomaly synthesis (SIAS) has emerged as a promising alternative; however, existing methods struggle to balance sampling efficiency and generation quality. Moreover, most approaches treat all spatial regions uniformly, overlooking the distinct statistical differences between anomaly and background areas. This uniform treatment hinders the synthesis of controllable, structure-specific anomalies tailored for segmentation tasks. In this paper, we propose FAST, a foreground-aware diffusion framework featuring two novel modules: the Anomaly-Informed Accelerated Sampling (AIAS) and the Foreground-Aware Reconstruction Module (FARM). AIAS is a training-free sampling algorithm specifically designed for segmentation-oriented industrial anomaly synthesis, which accelerates the reverse process through coarse-to-fine aggregation and enables the synthesis of state-of-the-art segmentation-oriented anomalies in as few as 10 steps. Meanwhile, FARM adaptively adjusts the anomaly-aware noise within the masked foreground regions at each sampling step, preserving localized anomaly signals throughout the denoising trajectory. Extensive experiments on multiple industrial benchmarks demonstrate that FAST consistently outperforms existing anomaly synthesis methods in downstream segmentation tasks. We release the code at: https://anonymous.4open.science/r/NeurIPS-938.

Problem

• Industrial anomaly segmentation lacks sufficient pixel-level anomaly annotations
• Existing methods fail to balance sampling efficiency and generation quality
• Current approaches overlook statistical differences between anomaly and background regions

Innovation

• Anomaly-Informed Accelerated Sampling for efficient generation
• Foreground-Aware Reconstruction Module for localized anomaly adjustment
• Coarse-to-fine aggregation enables synthesis in 10 steps