DualDiff

Official DualDiff repair checkpoint for DualDiff3D.

DualDiff repairs a degraded novel-view rendering using its structure together with the appearance of a nearby reference image.

Usage

The DualDiff3D code downloads this checkpoint automatically:

python -m DualDiff.hub

Run image-pair inference:

python DualDiff/infer.py \
  --input_image /path/to/degraded_render.png \
  --ref_image /path/to/reference.png \
  --output_dir outputs/repair

The checkpoint contains the complete structure and appearance U-Net state dictionaries plus the trained VAE LoRA and skip-connection parameters. It does not contain optimizer state. The SD-Turbo VAE, text encoder, tokenizer, scheduler, and base configuration are loaded separately by the code.

Citation

@inproceedings{wang2026dualdiff3d,
  title     = {DualDiff3D: Dual Structure-Appearance Diffusion
               Priors for Reliability-Enhanced 3D Gaussian Splatting},
  author    = {Qian Wang and Yu Wang and Weiqi Li and Xinhua Cheng
               and Xiandong Meng and Ronggang Wang and Jian Zhang},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year      = {2026}
}

License

The DualDiff code is released under Apache-2.0. This checkpoint is derived from SD-Turbo; use of the checkpoint is also subject to the applicable SD-Turbo model license.