Jia Qin, Youyi Zheng, and Kun Zhou. 2022. Motion In-betweening via Two-stage Transformers. ACM Trans. Graph. 41, 6, Article 184 (December 2022), 16 pages. https://doi.org/10.1145/3550454.3555454
Based on repository of above-mentioned paper: https://github.com/victorqin/motion_inbetweening Test on Python 3.11 on a laptop without GPU.
-
Download LAFAN1 dataset.
-
Extract
lafan1.zip
todatasets
folder. Bvh files should be located inmotion_inbetweening/datasets/lafan1
folder. -
Download the pre-trained models from the Releases Page. Extract it to the
motion_inbetweening/experiments
folder. -
Install PyTorch. The code has been tested in Python3.8, PyTorch-1.8.2.
To use the full method (Detail + Context Transformer) to generate in-betweening, run eval_detail_model.py
.
Usage:
usage: eval_detail_model.py [-h] [-o OFFSET] [-t TRANS] [-d] [-p]
det_config ctx_config animation output_dir
Evaluate detail model. No post-processing applied by default.
positional arguments:
det_config detail config name
ctx_config context config name
animation path to BVH file with animation
output_dir output directory where both reference animation and
inferred animation should be placed
options:
-h, --help show this help message and exit
-o OFFSET, --offset OFFSET
animation clip frame offset from start, should be >=
175 (default=175)
-t TRANS, --trans TRANS
transition length (default=30)
-d, --debug debug mode
-p, --post_processing
apply post-processing
Example:
Get inferred result on animation aiming1_subject1.bvh from LAFAN1 dataset with transition=30 frames,
where the animation clip starts at frame 450, save to the output
directory:
python eval_detail_model.py lafan1_detail_model lafan1_context_model "datasets/lafan1/aiming1_subject1.bvh" output -t 30 --offset 450