SimMotionEdit: Text-Based Human Motion Editing with Motion Similarity Prediction

Zhengyuan Li  Kai Cheng  Anindita Ghosh  Uttaran Bhattacharya  Liang-Yan Gui  Aniket Bera 
Purdue University, DFKI, MPI-INF,
Adobe Inc., University of Illinois Urbana-Champaign
CVPR 2025

Motion Similarity	Text-based Motion Editing

Environment Setup

Please follow motionfix to download the dataset and set up the environment.

We provide pretrained model and pre-processed similarity in this link.

Constant setup: set preproc.sim_file to the path of preprocessed similarity file in ./configs/data/motionfix.yaml.

Evaluation

Step 1: Extract the samples

python motionfix_evaluate.py folder=/path/to/exp/ guidance_scale_text_n_motion=2.0 guidance_scale_motion=2.0 data=motionfix

Step 2: Compute the metrics

python compute_metrics.py folder=/path/to/exp/samples/npys

Training

python -u train.py --config-name="train_cls_arch" experiment=cls_arch run_id=no_text

Acknowledgements

Our code is based on: motionfix.

License

This code is distributed under an MIT LICENSE. We also include the LICENSE of motionfix in our repo. Other third-party datasets and software are subject to their respective licenses.