DF3DV-1K-Fixer

Project Page | Paper | GitHub

The dataset provides masked paired 3DGS/Distractor-Free 3DGS rendering results and ground-truth images from DF3DV-1K. It is designed to facilitate research in distractor-free novel view synthesis and image-to-image restoration for 3D radiance fields.

Directory Structure

├── Train
│   └── DF3DV-1K-Star-Fixer
│       ├── 0000
│       │   ├── 040625-LundoBin
│       │   │   └── 040625-LundoBin-All
│       │   │       ├── MODELS
│       │   │       │   ├── 3DGS (masked rendering results and ground-truth images)
│       │   │       │   │   ├── 040625-LundoBin_3DGS_extra_IMG_6957_source.png
│       │   │       │   │   ├── ...
|       |   │       │   │   └── 040625-LundoBin_3DGS_extra_IMG_7041_target.png
│       │   │       │   ├── ...
│       │   │       │   └── WILDGS
│       │   │       ├── NN (nearest neighbors for images)
|       |   |       │   └── nn_rank_by_name.json
│       │   │       └── undistortion_sparse (COLMAP result)
│       │   │           └── 0
│       │   │               ├── cameras.bin
│       │   │               ├── cameras.txt
│       │   │               ├── images.bin
│       │   │               ├── images.txt
│       │   │               ├── points3D.bin
│       │   │               └── points3D.txt
│       │   ├── ...
│       │   └── 090625-BlueBikeBell
│       ├── ...
│       └── 0024
└── Val
    ├── DF3DV-41-Fixer
    │   ├── 021125-Chess
    │   │   └── 021125-Chess-All
    │   │       ├── MODELS
    │   │       │   ├── 3DGS
    │   │       │   ├── ...
    │   │       │   └── WILDGS
    │   │       ├── NN
    │   │       └── undistortion_sparse
    │   │           └── 0
    │   ├── ...   
    │   └── 301025-TempleDrumIncense
    └── on-the-go
        ├── corner
        │   └── corner-All
        │       ├── MODELS
        │       │   ├── 3DGS
        │       │   ├── ...
        │       │   └── WILDGS
        │       ├── NN
        │       └── undistortion_sparse
        │           └── 0
        ├── ...
        └── spot

Download Instruction

# Install the Hugging Face CLI
pip install -U "huggingface_hub[cli]"

# Login to your Hugging Face account
hf auth login

# Download whole dataset
# Example
hf download ChengYou305/DF3DV-1K-Fixer --repo-type dataset --local-dir DF3DV-1K-Fixer

# Download training set
# Example
hf download ChengYou305/DF3DV-1K-Fixer --repo-type dataset --local-dir DF3DV-1K-Fixer --include "Train/*"

# Download testing set
# Example
hf download ChengYou305/DF3DV-1K-Fixer --repo-type dataset --local-dir DF3DV-1K-Fixer --include "Val/*"

# Download specific chunk
# Example
hf download ChengYou305/DF3DV-1K-Fixer --repo-type dataset --local-dir DF3DV-1K-Fixer --include "Train/DF3DV-1K-Star-Fixer/0000/*" 

# Download everything except specific files
# Example
hf download ChengYou305/DF3DV-1K-Fixer --repo-type dataset --local-dir DF3DV-1K-Fixer --exclude "Mask.zip" 

More Information

Please refer this repo for more details.

Citation

@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}
}