3D-CustomBench

3D-CustomBench is the multi-view subject benchmark introduced in 3DreamBooth: High-Fidelity 3D Subject-Driven Video Generation Model. It contains 30 subjects with ordered multi-view captures, background-normalized reference images, and evaluation prompts for customized video generation.

Dataset summary

Item	Count
Subjects	30
Multi-view images	897
Reference images	122

Each subject provides full 360-degree visual coverage for evaluating subject fidelity and 3D geometric consistency.

Dataset structure

3D-CustomBench/
├── README.md
├── manifest.json
└── subjects/
    └── graduation_bear/
        ├── images/
        │   ├── 001.jpeg
        │   └── ...
        ├── references/
        │   ├── 001.png
        │   └── ...
        ├── metadata.json
        └── prompt.txt

• images/: ordered multi-view captures used for subject customization and evaluation.
• references/: background-normalized conditioning images used by 3Dapter and Joint.
• prompt.txt: subject-specific evaluation prompt.
• metadata.json: stable public ID, legacy ID, prompt, and file counts.
• manifest.json: index and metadata for all subjects.

Public subject IDs use descriptive snake_case. The legacy_id field is retained only to reproduce internal experiments.

Download

hf download lanikoworld/3D-CustomBench \
  --repo-type dataset \
  --local-dir ./datasets/3d-custombench

From the 3DreamBooth repository:

python scripts/data/download_custombench.py

3DreamBooth example

python scripts/run.py configs/examples/graduation_bear/train_joint.yaml
python scripts/run.py configs/examples/graduation_bear/validate_joint.yaml

License

3D-CustomBench is released under the Creative Commons Attribution 4.0 International License (CC BY 4.0). You may share and adapt the dataset for any purpose with appropriate attribution. This license covers only rights held by the dataset authors; third-party rights such as trademarks are not granted.

Citation

If you use 3D-CustomBench, please cite the 3DreamBooth paper:

@misc{ko20263dreambooth,
  title         = {3DreamBooth: High-Fidelity 3D Subject-Driven Video Generation Model},
  author        = {Hyun-kyu Ko and Jihyeon Park and Younghyun Kim and Dongheok Park and Eunbyung Park},
  year          = {2026},
  eprint        = {2603.18524},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  url           = {https://arxiv.org/abs/2603.18524}
}

• Paper
• Project page
• Code