SynSur: An end-to-end generative pipeline for synthetic industrial surface defect generation and detection

arXiv cs.AI / 4/30/2026

💬 OpinionIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper argues that industrial defect detection bottlenecks are often driven by scarce labeled defect data rather than model capacity, motivating synthetic data generation.
It proposes an end-to-end pipeline that uses vision-language-model prompts, LoRA-adapted diffusion, mask-guided inpainting, and automatic label derivation with sample filtering.
Experiments on ball screw drive pitting defects and cross-domain tests on the Mobile phone screen defect (MSD) segmentation dataset evaluate both defect detection performance and which pipeline stages produce realistic, useful samples.
Results with YOLOv26, YOLOX, and LW-DETR indicate that training solely on synthetic defects cannot replace real data, but combining synthetic and real data can preserve performance and sometimes provide modest gains.
The authors conclude the main value of diffusion-based synthetic defect synthesis is strengthening limited real datasets, with domain adaptation and annotation-quality control remaining critical for transfer.

Abstract

The bottleneck in learning-based industrial defect detection is often limited not by model capacity, but by the scarcity of labeled defect data: defects are rare, annotations are expensive, and collecting balanced training sets is slow. We present an end-to-end pipeline for synthetic defect generation and annotation, combining Vision-Language-Model-based prompts, LoRA-adapted diffusion, mask-guided inpainting, and sample filtering with automatic label derivation, and demonstrates the potential of real data with realistic synthetic samples to overcome data scarcity. The evaluation is conducted on, a challenging dataset of pitting defects on ball screw drives, and then on a subset of the Mobile phone screen surface defect segmentation dataset (MSD) dataset to test cross-domain transfer. Beyond downstream detector performance, we analyze key stages of the pipeline, including prompt construction, LoRA selection, and sample filtering with DreamSim and CLIPScore, to understand which synthetic samples are both realistic and useful. Experiments with YOLOv26, YOLOX, and LW-DETR show that synthetic-only training does not replace real data. When combined with real data, synthetic defects can preserve performance and yield modest gains in selected BSData training regimes. The MSD transfer study shows that the overall pipeline structure carries over to a second industrial inspection domain, while also highlighting the importance of domain-specific adaptation and annotation-quality control. Overall, the paper provides an end-to-end assessment of diffusion-based industrial defect synthesis and shows that its strongest value lies in strengthening scarce real datasets rather than substituting for them.