Dataset Viewer
Auto-converted to Parquet
Search is not available for this dataset
image
imagewidth (px)
256
3.84k
End of preview. Expand in Data Studio

Defect Spectrum Dataset

Welcome to the Defect Spectrum dataset repository. This comprehensive benchmark is a granular collection of large-scale defect datasets with rich semantics, designed to push the frontier of industrial defect inspection research and applications.

Paper: https://huggingface.co/papers/2310.17316

Github repository: https://github.com/EnVision-Research/Defect_Spectrum

Overview

Defect inspection is a critical component within the closed-loop manufacturing system. To facilitate advanced research and development in this domain, we introduce the Defect Spectrum dataset. It offers precise, semantics-abundant, and large-scale annotations for a wide range of industrial defects. This dataset is an enhancement over existing benchmarks, providing refined annotations and introducing detailed semantic layers, allowing for the distinction between multiple defect types within a single image.

Features

  • Semantics-Abundant Annotations: Each defect is meticulously labeled, not just at the pixel level but with rich contextual information, providing insights into the defect type and implications.
  • High Precision: Annotations are refined by experts to capture even the subtlest of defects, ensuring high precision.
  • Large-Scale Data: Building on four key industrial benchmarks, Defect Spectrum stands out with its extensive coverage and depth.
  • Incorporates Descriptive Captions: To bridge the gap towards Vision Language Models (VLMs), each sample is accompanied by a descriptive caption.

Directory Structure

DefectSpectrum/
β”œβ”€β”€ DS-MVTec/
β”‚   β”œβ”€β”€ bottle/
β”‚   β”‚   β”œβ”€β”€ image/         # Original images of the bottle category
β”‚   β”‚   β”œβ”€β”€ caption/       # Descriptive captions of the bottle category
β”‚   β”‚   β”œβ”€β”€ mask/          # Single channel defect masks for the bottle category
β”‚   β”‚   └── rgb_mask/      # Colored defect masks for better visualization
β”‚   β”œβ”€β”€ cable/
β”‚   β”‚   β”œβ”€β”€ image/         # Original images of the cable category
β”‚   β”‚   β”œβ”€β”€ caption/       # Descriptive captions of the cable category
β”‚   β”‚   β”œβ”€β”€ mask/          # Single channel defect masks for the cable category
β”‚   β”‚   └── rgb_mask/      # Colored defect masks for better visualization
β”‚   └── ...
β”œβ”€β”€ DS-VISION/
β”‚   └── ...
β”œβ”€β”€ DS-DAGM/
β”‚   └── ...
β”œβ”€β”€ DS-Cotton-Fabric/
β”‚   └── ...

To-Do List

  • Task 1: Release DS-MVTec image-mask pairs.
  • Task 2: Release DS-VISION, DS-DAGM, and DS-Cotton-Fabric image-mask pairs.
  • Task 3: Release captions.
  • Task 4: Release selected synthetic data.

license: mit

Downloads last month
2,927