Leveraging Latent Diffusion Models for Training-Free In-Distribution Data Augmentation for Surface Defect Detection

TL;DR

This paper introduces DIAG, a training-free, diffusion-based data augmentation method for defect detection that incorporates human guidance, improves over existing techniques, and achieves significant performance gains on the KSDD2 dataset.

Contribution

DIAG is a novel, training-free, diffusion-based in-distribution anomaly generation pipeline that uses human-in-the-loop guidance for improved defect data augmentation.

Findings

Achieves approximately 18% AP improvement with positive samples.

Achieves approximately 28% AP improvement without positive samples.

Operates in a zero-shot manner without fine-tuning.

Abstract

Defect detection is the task of identifying defects in production samples. Usually, defect detection classifiers are trained on ground-truth data formed by normal samples (negative data) and samples with defects (positive data), where the latter are consistently fewer than normal samples. State-of-the-art data augmentation procedures add synthetic defect data by superimposing artifacts to normal samples to mitigate problems related to unbalanced training data. These techniques often produce out-of-distribution images, resulting in systems that learn what is not a normal sample but cannot accurately identify what a defect looks like. In this work, we introduce DIAG, a training-free Diffusion-based In-distribution Anomaly Generation pipeline for data augmentation. Unlike conventional image generation techniques, we implement a human-in-the-loop pipeline, where domain experts provide multimodal guidance to the model through text descriptions and region localization of the possible anomalies. This strategic shift enhances the interpretability of results and fosters a more robust human feedback loop, facilitating iterative improvements of the generated outputs. Remarkably, our approach operates in a zero-shot manner, avoiding time-consuming fine-tuning procedures while achieving superior performance. We demonstrate the efficacy and versatility of DIAG with respect to state-of-the-art data augmentation approaches on the challenging KSDD2 dataset, with an improvement in AP of approximately 18% when positive samples are available and 28% when they are missing. The source code is available at https://github.com/intelligolabs/DIAG.