What is equivariant neural network design in computer vision?

Equivariant networks maintain structured transformations when inputs are transformed (e.g., rotation, translation).

Equivariance means that if the input is transformed, the output transforms in a predictable way. For example, rotating an image should rotate feature maps accordingly. Group equivariant CNNs enforce symmetry constraints using group theory, improving generalization and reducing sample complexity in vision tasks.