Skip to content

ACKNOWLEDGEMENT.md

Latest commit

 

History

History
25 lines (18 loc) · 948 Bytes

ACKNOWLEDGEMENT.md

File metadata and controls

25 lines (18 loc) · 948 Bytes

Acknowledgements

This release builds on the following open-source projects. We thank their authors.

CIPS

advimman/CIPS (Conditionally-Independent Pixel Synthesis). The neural field backbone in neural_field/model derives from CIPS.

HyperDiffusion

Rgtemze/HyperDiffusion. The weight_space_diffusion repo builds on HyperDiffusion: the diffusion-over-weights training and sampling pipeline, the weight atomizer/tokenizer, and the generation metric harness. HyperDiffusion is also the prior weight-space baseline we compare against (Tables 2 to 3).

DeepSDF

facebookresearch/DeepSDF. The Stage 1 base field is trained as a variational autodecoder following the DeepSDF formulation: a shared decoder with per-instance latent codes optimized jointly, applied here to 2D images and 3D occupancy/SDF fields.