MotionCritic

Pretrained critic model for "Aligning Human Motion Generation with Human Perceptions" (ICLR 2025).

[Paper] [GitHub] [Project Page]

Usage

from lib.model.load_critic import load_critic
from parsedata import into_critic
import torch

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
critic_model = load_critic("motioncritic_pre.pth", device)

# motion: [bs, 25, 6, frame], rot6d
preprocessed = into_critic(motion)  # [bs, frame, 25, 3], axis-angle
scores = critic_model.module.batch_critic(preprocessed)

Citation

@inproceedings{motioncritic2025,
    title={Aligning Motion Generation with Human Perceptions},
    author={Wang, Haoru and Zhu, Wentao and Miao, Luyi and Xu, Yishu and Gao, Feng and Tian, Qi and Wang, Yizhou},
    booktitle={International Conference on Learning Representations (ICLR)},
    year={2025}
}