File size: 2,985 Bytes
7ddb28d |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 |
# DiagramQG Dataset
*Figure 1: Four different examples of different subjects in DiagramQG dataset.*
*Figure 2: Domain diversity in DiagramQG. Each color corresponds to one subject: Natural Science (blue), Earth Science (yellow), Applied Science (green), and Social Science (orange).*
## Overview
DiagramQG is a comprehensive educational dataset focused on scientific diagram question generation. It contains:
- 19,475 unique questions
- 8,372 diagrams
- 44,472 combinations of (target & concept text constraint, diagram, question)
- Coverage across 4 subjects, 15 courses, and 169 concepts
## Dataset Structure
### Subject Areas
The dataset covers four main subject areas:
- Natural Science
- Earth Science
- Applied Science
- Social Science
### Hierarchical Organization
Data is organized hierarchically:
1. Subject (e.g., Natural Science)
2. Course (e.g., Biology)
3. Concept (e.g., Ecological interactions)
## Data Collection Process
### Phase 1: Initial Data Gathering
- Sources: Existing datasets and Google Image Search
- Raw dataset: 20,000+ diagrams and 40,000+ questions
### Phase 2: Organization
- Classification into 4 subjects and 15 courses
- Mapping questions to 169 distinct concepts
### Phase 3: Annotation
- Trained crowd workers annotate:
- Target & concept text constraints
- Diagram elements and texts
- Produced 70,000+ unique combinations
### Phase 4: Quality Assurance
- Secondary crowd worker evaluation (0-100 scale)
- Filtered combinations below 60 points
- Final dataset: 44,472 validated combinations
## Dataset Analysis
### Question Distribution
*Figure 3: Question distribution in DiagramQG.*
### Concept Distribution
*Figure 4: Distribution of diagrams, questions, and questions per diagram ratios across different concepts in DiagramQG.*
### Dataset Comparison
| Dataset | Questions | Images | Objects/Image | Image Type | Constraints | Knowledge Type |
|---------|-----------|---------|---------------|------------|-------------|----------------|
| VQAv2.0 | 1.1M | 20k | 3.5 | natural | answer | N/A |
| FVQA | 5,826 | 2k | 2.9 | natural | answer | common-sense |
| VQG-COCO | 25,000 | 5k | 3.3 | natural | image, caption | common-sense |
| K-VQG | 16,098 | 13K | 2.7 | natural | knowledge triple | common-sense |
| DiagramQG | 19,475 | 8,372 | 11.2 | diagram | target, concept | subject knowledge |
## Unique Challenges
1. **Domain-specific Knowledge Requirement**
- Requires understanding of specialized subject concepts
- Goes beyond common sense reasoning
2. **Long-tail Distribution**
- Uneven concept coverage
- Challenges in model generalization
3. **High Information Density**
- Complex diagram interpretation
- Dense visual information processing |