Datasets:
Modalities:
Image
Formats:
imagefolder
Languages:
English
Size:
10K - 100K
ArXiv:
Tags:
industry
License:
DefectSpectrum
commited on
Commit
Β·
e70d490
1
Parent(s):
cdb0c8f
readme
Browse files
README.md
ADDED
@@ -0,0 +1,62 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: mit
|
3 |
+
task_categories:
|
4 |
+
- image-segmentation
|
5 |
+
- image-to-text
|
6 |
+
language:
|
7 |
+
- en
|
8 |
+
tags:
|
9 |
+
- industry
|
10 |
+
pretty_name: DefectSpectrum
|
11 |
+
size_categories:
|
12 |
+
- 1K<n<10K
|
13 |
+
---
|
14 |
+
# Defect Spectrum Dataset
|
15 |
+
|
16 |
+
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.
|
17 |
+
|
18 |
+
|
19 |
+
## Overview
|
20 |
+
|
21 |
+
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.
|
22 |
+
|
23 |
+
### Features
|
24 |
+
|
25 |
+
- **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.
|
26 |
+
- **High Precision**: Annotations are refined by experts to capture even the subtlest of defects, ensuring high precision.
|
27 |
+
- **Large-Scale Data**: Building on four key industrial benchmarks, Defect Spectrum stands out with its extensive coverage and depth.
|
28 |
+
- **Incorporates Descriptive Captions**: To bridge the gap towards Vision Language Models (VLMs), each sample is accompanied by a descriptive caption.
|
29 |
+
|
30 |
+
### Directory Structure
|
31 |
+
|
32 |
+
```plaintext
|
33 |
+
DefectSpectrum/
|
34 |
+
βββ DS-MVTec/
|
35 |
+
β βββ bottle/
|
36 |
+
β β βββ image/ # Original images of the bottle category
|
37 |
+
β β βββ caption/ # Descriptive captions of the bottle category
|
38 |
+
β β βββ mask/ # Single channel defect masks for the bottle category
|
39 |
+
β β βββ rgb_mask/ # Colored defect masks for better visualization
|
40 |
+
β βββ cable/
|
41 |
+
β β βββ image/ # Original images of the cable category
|
42 |
+
β β βββ caption/ # Descriptive captions of the cable category
|
43 |
+
β β βββ mask/ # Single channel defect masks for the cable category
|
44 |
+
β β βββ rgb_mask/ # Colored defect masks for better visualization
|
45 |
+
β βββ ...
|
46 |
+
βββ DS-VISION/
|
47 |
+
β βββ ...
|
48 |
+
βββ DS-DAGM/
|
49 |
+
β βββ ...
|
50 |
+
βββ DS-Cotton-Fabric/
|
51 |
+
β βββ ...
|
52 |
+
```
|
53 |
+
|
54 |
+
## To-Do List
|
55 |
+
- [ ] Task 1: Release DS-MVTec image-mask pairs.
|
56 |
+
- [ ] Task 2: Release DS-VISION, DS-DAGM, and DS-Cotton-Fabric image-mask pairs.
|
57 |
+
- [ ] Task 3: Release captions.
|
58 |
+
- [ ] Task 4: Release selected synthetic data.
|
59 |
+
|
60 |
+
---
|
61 |
+
license: mit
|
62 |
+
---
|