Skip to content

taherahmadi/STPOTR

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

43 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

STPOTR

This is a repository for the paper "STPOTR: Simultaneous Human Trajectory and Pose Prediction Using a Non-Autoregressive Transformer for Robot Following Ahead". It includes the codes for Simultaneous Human Trajectory and Pose Prediction training and testing. This work is partially inspired by the potr paper.

alt text

Resulted Motions

Left=Model Prediction / Right = Ground Truth

Blue frames are input and red frames are output alt text

Installation

conda create --name stpotr --file stpotr.yml
conda activate stpotr
pip install numpy
pip install opencv-python

DATA

We have trained the model with Human3.6m dataset. You need to download this data file and move it to the data folder.

Testing

We have provided a pretrained model. You can download this folder and move it to the main repo folder.

You can use this command to run the model.

cd model
python testing.py

Training

In order to train the model you can run this command:

cd training
python transformer_model_fn.py --model_prefix=./trained_model \
--batch_size=16 \
--data_path=../data/h3.6m/ \
--learning_rate=0.0001 \
--max_epochs=300 \
--steps_per_epoch=200 \
--loss_fn=mse \
--model_dim=512 \
--model_dim_traj=64 \
--num_encoder_layers=4 \
--num_decoder_layers=4 \
--num_heads=8 \
--dim_ffn=2048 \
--dropout=0.3 \
--lr_step_size=200 \
--gamma=0.1 \
--learning_rate_fn=step \
--pose_format=expmap \
--pose_embedding_type=gcn_enc \
--dataset=h36m_v3 \
--pre_normalization \
--pad_decoder_inputs \
--non_autoregressive \
--pos_enc_alpha=10 \
--pos_enc_beta=500 \
--action=all \
--warmup_epochs=50 \
--weight_decay=0.00001 \
--include_last_obs

Citation

If you happen to use the code for your research, please cite the following paper

@article{mahdavian2022stpotr,
  title={STPOTR: Simultaneous Human Trajectory and Pose Prediction Using a Non-Autoregressive Transformer for Robot Following Ahead},
  author={Mahdavian, Mohammad and Nikdel, Payam and TaherAhmadi, Mahdi and Chen, Mo},
  journal={arXiv preprint arXiv:2209.07600},
  year={2022}
}

About

Human Pose and Hip Trajectory Prediction Using Transformers

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages

  • Python 100.0%