| pipeline_tag: image-to-image | |
| tags: | |
| - Novel View Synthesis | |
| - NeRF | |
| - 3D Gaussian Splatting | |
| - Diffusion Models | |
| - Image Restoration | |
| - Fixer | |
| - Distractor | |
| # DI<sup>2</sup>FIX | |
| [**Project Page**](https://johnnylu305.github.io/df3dv1k_web/) | [**Paper**](https://huggingface.co/papers/2604.13416) | [**GitHub**](https://github.com/johnnylu305/DF3DV) | [**Demo**](https://chengyou305-di2fix-demo.hf.space/) | |
| DI<sup>2</sup>FIX (Distractor-Free DIFIX) is a plug-and-play 2D enhancer designed to improve radiance field renderings (such as 3D Gaussian Splatting and NeRF) by removing distractors and artifacts. It was introduced in the paper [DF3DV-1K: A Large-Scale Dataset and Benchmark for Distractor-Free Novel View Synthesis](https://huggingface.co/papers/2604.13416). | |
| ## Method Overview | |
| DI<sup>2</sup>FIX is a diffusion-based 2D enhancer fine-tuned on the DF3DV-1K dataset. It refines and enhances initial novel-view synthesis outputs, removing distractors and significantly improving rendering quality (achieving average improvements of 0.96 dB PSNR and 0.057 LPIPS on the held-out DF3DV-41 benchmark and the On-the-go dataset). | |
| ## Citation | |
| ```bibtex | |
| @article{lu2026df3dv, | |
| title={DF3DV-1K: A Large-Scale Dataset and Benchmark for Distractor-Free Novel View Synthesis}, | |
| author={Lu, Cheng-You and Hung, Yi-Shan and Chi, Wei-Ling and Wang, Hao-Ping and Tsai, Charlie Li-Ting and Chang, Yu-Cheng and Liu, Yu-Lun and Do, Thomas and Lin, Chin-Teng}, | |
| journal={arXiv preprint arXiv:2604.13416}, | |
| year={2026} | |
| } | |
| ``` |