DF3DV-1K: A Large-Scale Dataset and Benchmark for Distractor-Free Novel View Synthesis

Project Page | Paper | GitHub

DF3DV-1K is a large-scale real-world dataset comprising 1,048 scenes, each providing clean and cluttered image sets for benchmarking distractor-free radiance fields. In total, the dataset contains 89,924 images captured using consumer cameras to mimic casual capture, spanning 128 distractor types and 161 scene themes across indoor and outdoor environments.

Directory Structure

├── DF3DV-1K-Star
│   ├── 0000
│   │   ├── 040625-LundoBin
│   │   │   ├── 040625-LundoBin-All (curated data)
│   │   │   │   ├── images (COLMAP input images)
│   │   │   │   │   ├── clutter_IMG_7042.JPG 
│   │   │   │   │   ├── ...
│   │   │   │   │   └── extra_IMG_7041.JPG
│   │   │   │   ├── sparse (COLMAP result)
│   │   │   │   │   └── 0
│   │   │   │   │       ├── cameras.bin
│   │   │   │   │       ├── images.bin
│   │   │   │   │       ├── points3D.bin
│   │   │   │   │       └── project.ini
│   │   │   │   ├── split.json (list of clean and cluttered images)
│   │   │   │   ├── transforms_clutter.json (Instant-NGP JSON file for cluttered images only)
│   │   │   │   ├── transforms_extra.json (Instant-NGP JSON file for clean images only)
│   │   │   │   ├── transforms.json (Instant-NGP JSON file for all images)
│   │   │   │   ├── undistortion_images (COLMAP-undistorted images)
│   │   │   │   │   ├── clutter_IMG_7042.JPG
│   │   │   │   │   ├── ...
│   │   │   │   │   └── extra_IMG_7041.JPG
│   │   │   │   └── undistortion_sparse (COLMAP-undistorted result)
│   │   │   │       └── 0
│   │   │   │           ├── cameras.bin
│   │   │   │           ├── cameras.txt
│   │   │   │           ├── images.bin
│   │   │   │           ├── images.txt
│   │   │   │           ├── points3D.bin
│   │   │   │           └── points3D.txt
│   │   │   ├── 040625-LundoBin-Clean (candidate clean images)
│   │   │   │   └── images
│   │   │   │       ├── IMG_6957.JPG
│   │   │   │       ├── ...
│   │   │   │       └── IMG_7041.JPG
│   │   │   └── 040625-LundoBin-Clutter (candidate cluttered images)
│   │   │       └── images
│   │   │            ├── IMG_7042.JPG
│   │   │            ├── ...
│   │   │            └── IMG_7140.JPG
│   │   ├── ...
│   │   └── 090625-BlueBikeBell
│   ├── ...
│   └── 0024
└── DF3DV-41
    ├── 021125-Chess
    │   ├── 021125-Chess-All
    │   │   ├── images
    │   │   ├── sparse
    │   │   │   └── 0
    │   │   ├── undistortion_images
    │   │   └── undistortion_sparse
    │   │       └── 0
    │   ├── 021125-Chess-Clean
    │   │   └── images
    │   └── 021125-Chess-Clutter
    │       └── images
    ├── ...
    └── 301025-TempleDrumIncense

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
hf download ChengYou305/DF3DV-1K --repo-type dataset --local-dir DF3DV-1K

# Download DF3DV-1K*
hf download ChengYou305/DF3DV-1K --repo-type dataset --local-dir DF3DV-1K --include "DF3DV-1K-Star/*"

# Download DF3DV-41
hf download ChengYou305/DF3DV-1K --repo-type dataset --local-dir DF3DV-1K --include "DF3DV-41/*"

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

# Download everything except specific files
hf download ChengYou305/DF3DV-1K --repo-type dataset --local-dir DF3DV-1K --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}
}