Group-UNet

• Testing equivariance in neural networks using Group Convolutions applied to segmentation of Butterfly images
• Implementation of U-Net using Group and Lifting Convolutions
• Logging and hyper-parameter sweeps with weights and biases
• Utilities to evaluate equivariance under rotation
• Comparison with data augmentation techniques to achieve similar results
• Dataset can be found here

Josiah Wang, Katja Markert, and Mark Everingham
Learning Models for Object Recognition from Natural Language Descriptions
In Proceedings of the 20th British Machine Vision Conference (BMVC2009)