A dataset generator for testing rotation equivariance in neural networks, specifically focusing on segmentation tasks using simple geometric shapes.
- Image Size: 128 x 128 pixels
- Number of Samples: 100 images
- Rotation Range: 0° to 360° (uniformly distributed)
- Shape Type: Ellipses with major axis 40 pixels and minor axis 20 pixels, Arrows
python generate.py