• This repo is official implementation for mmTransformer in pytorch. Currently, the core code of mmTransformer is implemented in the commercial project, we provide inference code of model with six trajectory propopals for your reference.

• For other information, please refer to our paper Multimodal Motion Prediction with Stacked Transformers. (CVPR 2021) [Paper] [Webpage]

• Initialize virtual environment:

  conda create -n mmTrans python=3.7
  

• Install agoverse api. Please refer to this page.

• Install the pytorch. The latest codes are tested on Ubuntu 16.04, CUDA11.1, PyTorch 1.8 and Python 3.7: (Note that we require the version of torch >= 1.5.0 for testing with pretrained model)

  pip install torch==1.8.0+cu111\
        torchvision==0.9.0+cu111\
        torchaudio==0.8.0 -f https://download.pytorch.org/whl/torch_stable.html
  

• For other requirement, please install with following command:

  pip install -r requirement.txt
  

1. Clone this repo from the GitHub.

    git clone https://github.com/decisionforce/mmTransformer.git
   

2. Download the pretrained model and data [here] (map.pkl for Python 3.7 is available [here]) and save it to ./models and ./interm_data.

    cd mmTransformer
    mkdir models
    mkdir interm_data
   

3. Finally, your directory structure should look something like this:

    mmTransformer
    └── models
        └── demo.pt
    └── interm_data
        └── argoverse_info_val.pkl
        └── map.pkl
   

Alternatively, you can process the data from scratch using following commands.

1. Download Argoverse dataset and create a symbolic link to ./data folder or use following commands.

    cd path/to/mmtransformer/root
    mkdir data
    cd data
    wget https://s3.amazonaws.com/argoai-argoverse/forecasting_val_v1.1.tar.gz 
    tar -zxvf  forecasting_val_v1.1.tar.gz
   

2. Then extract the agent and map information from raw data via Argoverse API:

    python -m lib.dataset.argoverse_convertor ./config/demo.py
   

3. Finally, your directory structure should look something like above illustrated.

Format of processed data in ‘argoverse_info_val.pkl’:

Format of map information in ‘map.pkl’:

For testing:

python Evaluation.py ./config/demo.py --model-name demo

Here we showcase the expected results on validation set:

• We are going to open source our visualization tools and a demo result. (TBD)

If you have any issues with the code, please contact to this email: [email protected]

If you find our work useful for your research, please consider citing the paper

@article{liu2021multimodal,
  title={Multimodal Motion Prediction with Stacked Transformers},
  author={Liu, Yicheng and Zhang, Jinghuai and Fang, Liangji and Jiang, Qinhong and Zhou, Bolei},
  journal={Computer Vision and Pattern Recognition},
  year={2021}
}