Code of the paper "Exploring Vision Transformers for 3D Human Motion-Language Models with Motion Patches" (CVPR 2024).

• Python 3.11
• PyTorch 2.0.1+

poetry install

conda create -n MoPa python=3.11
conda activate MoPa
conda install pytorch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 pytorch-cuda=11.8 -c pytorch -c nvidia
pip install -r requirements.txt

Download HumanML3D Dataset and KIT-ML Dataset from the repository of HumanML3D. Unzip and locate them in the data folder.

The whole directory should be look like this:

MotionPatches
│   README.md
│   requirements.txt
|   ...
|
└───conf
└───scripts
└───...
│   
└───data
    └───HumanML3D
    |   └───new_joint_vecs
    |   └───new_joints
    |   └───...
    │   
    └───KIT-ML
        └───new_joint_vecs
        └───new_joints
        └───...

Then calculate the mean and variance of each dataset by:

python scripts/cal_mean_var.py

Download pre-trained model from huggingface and put them in checkpoints/pretrained/.

Evaluate the model with HumanML3D via retrieval:

python scripts/test.py dataset=HumanML3D exp_name=pretrained

Evaluate the model with KIT-ML via retrieval:

python scripts/test.py dataset=KIT-ML exp_name=pretrained

Using scripts/test_batch.py can get the result of small batches with 32 samples.

Train the model with HumanML3D:

python scripts/train.py dataset=HumanML3D

Train the model with KIT-ML:

python scripts/train.py dataset=KIT-ML

Evaluate the model with HumanML3D via retrieval:

python scripts/test.py dataset=HumanML3D

Evaluate the model with KIT-ML via retrieval:

python scripts/test.py dataset=KIT-ML

Using scripts/test_batch.py can get the result of small batches with 32 samples.

@InProceedings{yu2024exploring,
  title={Exploring Vision Transformers for 3D Human Motion-Language Models with Motion Patches},
  author={Yu, Qing and Tanaka, Mikihiro and Fujiwara, Kent},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2024}
}

CC BY-NC 4.0

Additionally, this repository contains third-party software. Refer NOTICE.txt for more details and follow the terms and conditions of their use.