DeformMLP:Dynamic Large-scale Receptive Field MLP Networks for Human Motion Prediction

Official PyTorch Implementation of the paper: DeformMLP:Dynamic Large-scale Receptive Field MLP Networks for Human Motion Prediction.

Haitao Huang, Chi-Man Pun, Haolun Li, Mengqi Liu, Jian Xiong , Hao Gao

Installation

To setup the environment:

cd DeformMLP
conda create -n DeformMLP python=3.6.13
conda activate DeformMLP
pip install -r requirements.txt

Data

Due to licensing it is not possible to provide any data. Please refer to STSGCN for the preparation of the dataset files.

Training

To train the model on h36m or amass, you can use the following commands:

python h36m/train_mixer_h36m.py --input_n 10 --output_n 25 --skip_rate 1 

python 3dpw/train_mixer_3dpw.py --input_n 10 --output_n 25 --skip_rate 5 

Evaluation

To test the pretrained models, you can use the following commands:

python h36m/test_mixer_h36m.py --input_n 10 --output_n 25 --skip_rate 1 

python 3dpw/test_mixer_3dpw.py --input_n 10 --output_n 25 --skip_rate 5 

Models

We release the pretrained models for academic purpose. You can download them from Google Drive. Unzip the .zip file in the /checkpoints directory.

Acknowledgments

Some of our code was adapted from HisRepsItself, MotionMixer and STSGCN. We thank the authors for making their code public.