Equivariant Few-Shot Generation of Implicit Neural Representation Weights
Generating diverse yet functionally consistent implicit neural representation (INR) weights from only a few examples. EQUIGEN exploits weight-space symmetry — functionally similar networks related by neuron permutations form an equivariance group — via an equivariant encoder trained with contrastive learning and smooth augmentation, an equivariance-guided diffusion process, and controlled perturbations in the equivariant subspace, validated on 2D image and 3D shape INR datasets (arXiv:2501.01601, cs.CV).
Few-shot Implicit Function Generation via Equivariance
Implicit Neural Representations (INRs) have emerged as a powerful framework for representing continuous signals. However, generating diverse INR weights remains challenging due to limited training da…