HMNeRFBench: A Comprehensive Benchmark for Neural Human Radiance Fields

This repository contains the implementation of the following paper:

A Comprehensive Benchmark for Neural Human Radiance Fields
Kenkun Liu¹², Derong Jin², Ailing Zeng¹, Xiaoguang Han², Lei Zhang^1
1International Digital Economy Academy ²The Chinese University of Hong Kong, Shenzhen

Datasets

1. ZJU-MoCap
2. GeneBody
3. HuMMan

Unifying Settings

Comparisons of recent NeRF-based human rendering methods on different aspects. In the column Dataset, ZM, PS, GB, HM, H36M, RP are ZJU-MoCap, People-Snapshot, GeneBody, HuMMan, Human3.6M, RenderPeople datasets, respectively. Views: train views for scene-specific methods and input views (*) for generalizable methods. Frames: train frames for scene-specific methods and input frames (*) for generalizable methods

Method	Dataset	Views	Frames	Generalizable	Animatable
NeuralBody	ZM, PS	4	100-300	✘	✔
AniNeRF	ZM, H36M	4	100-300	✘	✔
ARAH	ZM, H36M	4	300-400	✘	✔
HumanNeRF	ZM	1	500-600	✘	✔
UV-Volume	ZM, H36M	18	100	✘	✔
MonoHuman	ZM	1	500-600	✘	✔
NHP	ZM, AIST++	3*	1* or 3*	✔	✘
MPS-NeRF	ZM, H36M, THuman	3*	1*	✔	✔
GP-NeRF	ZM	3*	1*	✔	✘
KeypointNeRF	ZM	3*	1*	✔	✘
GNR	ZM, GB, RP	4*	1*	✔	✘

Benchmarking Scene-specific Methods (Coming soon...)

Benchmarking Generalizable Methods (Coming soon...)

GeneHumanNeRF: Animatable and Generalizable Human NeRF from Monocular Videos

🤝 Citation

If you find this repository useful for your work, please consider citing it as follows:

@inproceedings{liu2023comprehensive,
        title={A Comprehensive Benchmark for Neural Human Radiance Fields},
        author={Liu, Kenkun and Jin, Derong and Zeng, Ailing and Han, Xiaoguang and Zhang, Lei},
        booktitle={Thirty-seventh Conference on Neural Information Processing Systems Datasets and Benchmarks Track},
        year={2023}
      }