Abstract:Defect inspection is paramount within the closed-loop manufacturing system. However, existing datasets for defect inspection often lack precision and semantic granularity required for practical applications. In this paper, we introduce the Defect Spectrum, a comprehensive benchmark that offers precise, semantic-abundant, and large-scale annotations for a wide range of industrial defects. Building on four key industrial benchmarks, our dataset refines existing annotations and introduces rich semantic details, distinguishing multiple defect types within a single image. Furthermore, we introduce Defect-Gen, a two-stage diffusion-based generator designed to create high-quality and diverse defective images, even when working with limited datasets. The synthetic images generated by Defect-Gen significantly enhance the efficacy of defect inspection models. Overall, The Defect Spectrum dataset demonstrates its potential in defect inspection research, offering a solid platform for testing and refining advanced models.

What problem does this paper attempt to address?

### Problems Addressed by the Paper The paper primarily addresses several key issues in industrial defect detection: 1. **Limitations of Existing Datasets**: - Current datasets used for defect detection (such as DAGM, MVTEC, VISION, etc.) lack sufficient precision and semantic details in the annotation of defect types and locations. For example, these datasets typically provide only binary masks, which cannot distinguish between multiple types of defects. 2. **Introduction of a Comprehensive Benchmark Dataset**: - The authors introduce a comprehensive benchmark dataset named "Defect Spectrum," which provides precise, rich, and large-scale industrial defect annotations. These annotations not only refine existing defect categories but also offer detailed annotations of multiple defect types within the same image. With this dataset, researchers can conduct more comprehensive and precise defect detection analyses. 3. **Improving Model Performance**: - In industrial simulation experiments, using the Defect Spectrum dataset significantly improved the model's recall rate (increased by 10.74%) and reduced the false positive rate (decreased by 33.10%). Additionally, the paper proposes a generator based on a diffusion model called Defect-Gen, which can generate high-quality and diverse defect images under limited defect sample conditions, thereby significantly enhancing the performance of defect segmentation models (mIoU score increased by up to 9.85). 4. **Auxiliary Annotation Tool**: - An interactive annotation tool named "Defect-Click" is proposed to accelerate and simplify the process of fine defect annotation. This tool leverages pre-trained knowledge to automatically segment defect areas, significantly improving annotation efficiency. In summary, the paper aims to improve the performance of industrial defect detection tasks by introducing more detailed and rich datasets and generative models.

Defect Spectrum: A Granular Look of Large-Scale Defect Datasets with Rich Semantics

Saliency-Based Defect Detection in Industrial Images by Using Phase Spectrum

Coarse-to-Fine Few-Shot Defect Recognition with Dynamic Weighting and Joint Metric

A PRISMA Driven Systematic Review of Publicly Available Datasets for Benchmark and Model Developments for Industrial Defect Detection

Segment Anything Model for Industrial Vision: A Comprehensive Evaluation with a New Metric, a New Dataset, and a Toolbox

Automated Detection of Defects with Low Semantic Information in X-ray Images Based on Deep Learning

Defect classification for specular surfaces based on deflectometry and multi-modal fusion network

Interactive Defect Segmentation in X-Ray Images Based on Deep Learning

Defect-GAN: High-Fidelity Defect Synthesis for Automated Defect Inspection

ZJU-Leaper: A Benchmark Dataset for Fabric Defect Detection and a Comparative Study

Segment Anything in Defect Detection

A Casting Surface Dataset and Benchmark for Subtle and Confusable Defect Detection in Complex Contexts

Synth4Seg -- Learning Defect Data Synthesis for Defect Segmentation using Bi-level Optimization

Few-shot defect detection using feature enhancement and image generation for manufacturing quality inspection

An Adaptive Image Segmentation Network for Surface Defect Detection

High-resolution infrastructure defect detection dataset sourced by unmanned systems and validated with deep learning

Performance, Successes and Limitations of Deep Learning Semantic Segmentation of Multiple Defects in Transmission Electron Micrographs

Synthetic Defect Generation for Display Front-of-Screen Quality Inspection: A Survey

An Overview of Image Generation of Industrial Surface Defects

DDSNet: Deep Dual-Branch Networks for Surface Defect Segmentation

Addressing Class Imbalance and Data Limitations in Advanced Node Semiconductor Defect Inspection: A Generative Approach for SEM Images