Generates synthetic tasks showing a visual analogy A:B :: C:? with shape rotation transformations. The goal is to infer the rotation from A to B and apply the same rotation to C.
Each sample pairs a task (first frame + prompt describing what needs to happen) with its ground truth solution (final frame showing the result + video demonstrating how to achieve it). This structure enables both model evaluation and training.
| Property | Value |
|---|---|
| Task ID | O-8 |
| Task | Shape Rotation |
| Category | Abstraction |
| Resolution | 1024×1024 px |
| FPS | 16 fps |
| Duration | ~4.5 seconds |
| Output | PNG images + MP4 video |
# Clone the repository
git clone https://github.com/VBVR-DataFactory/O-8_shape_rotation_data-generator.git
cd O-8_shape_rotation_data-generator
# Install dependencies
pip install -r requirements.txt# Generate 100 samples
python examples/generate.py --num-samples 100
# Generate with specific seed
python examples/generate.py --num-samples 100 --seed 42
# Generate without videos
python examples/generate.py --num-samples 100 --no-videos
# Custom output directory
python examples/generate.py --num-samples 100 --output data/my_output| Argument | Type | Description | Default |
|---|---|---|---|
--num-samples |
int | Number of samples to generate | 100 |
--seed |
int | Random seed for reproducibility | Random |
--output |
str | Output directory | data/questions |
--no-videos |
flag | Skip video generation | False |
The image shows a visual analogy A:B :: C:? using simple shapes. A and C are the original shapes, B shows shape A after a rotation, and ? is the unknown result. Each shape has a small black marker dot to indicate its orientation. Your task is to infer the single rotation that maps A (pentagon) to B, then apply the same rotation to C (chevron) around its geometric center. The answer at ? must be shape C rotated by this angle, with size and position unchanged, so that the rotational relationship between A and B is mirrored between C and ?.
 |
 |
 |
| Initial Frame Analogy A:B :: C:? with rotation rule |
Animation Shape C rotates by same angle as A→B |
Final Frame Rotation rule applied to C |
Infer the rotation transformation from A to B in the analogy, then apply the same rotation to shape C to determine the answer.
- Analogy Format: A:B :: C:? with four positions
- Top Row (Rule): Shows shape A and its rotated version B
- Bottom Row (Query): Shows shape C and unknown answer ?
- Rotation Inference: Determine rotation angle from A to B - this generator modifies the rotation parameter of shapes while keeping all other properties (shape type, color, size, position) unchanged
- Rule Application: Apply same rotation to C around its geometric center
- Shape Preservation: C's size and position remain unchanged, only orientation changes
- Analogical reasoning: Understanding A:B :: C:? relationship pattern
- Rotation inference: Extracting rotation angle from example transformation
- Rule application: Applying inferred rotation to new shape
- Geometric center: Rotating around shape's center point
- Angle preservation: Maintaining exact rotation angle across different shapes
- Orientation markers: Each shape has a small black dot to indicate its orientation, making rotations visible even for symmetric shapes
- Abstract reasoning: Generalizing rotation transformation across different geometries
- Large-scale diversity: 20 distinct shapes, 46 rotation angles, and 400 colors for extensive variety (7M+ unique combinations)
data/questions/shape_rotation_task/shape_rotation_00000000/
├── first_frame.png # Initial state (analogy with query)
├── final_frame.png # Goal state (query shape rotated)
├── prompt.txt # Task instructions
├── ground_truth.mp4 # Solution video (16 fps)
└── question_metadata.json # Task metadata
File specifications: Images are 1024×1024 PNG. Videos are MP4 at 16 fps, approximately 4.5 seconds long.
analogical-reasoning rotation-transformation visual-analogy angle-inference geometric-transformation abstract-reasoning pattern-matching