Yiwei Lin; Hu Chuan; Baixuan Zhao; Hao Jiang; Yonghang Shan; Taojun Ding; Xi Zhang*

This repo contains the code for our paper:Anchor-based Multi-modal Transformer Network for Pedestrian Trajectory and Intention Prediction. The paper is accepted by CVCI2023 conference.

The code is mainly based on bitrap implementation.

Our network architecture:

Usage：

1. Download the dataset from JAAD official website.
2. Clone the repo and install the requirements.
3. Change the dataset path in configs/gtppo_tr_JAAD.yml
4. Run the code:

python tools/train.py --config_file configs/gtppo_tr_JAAD.yml
# or
python tools/test.py --config_file configs/gtppo_tr_JAAD.yml --CKPT_DIR your_weights.pth

5. if set VISUALIZATION: True and USE_WANDB: True in configs/gtppo_tr_JAAD.yml, the results will be visualized. ffmpeg and full dataset videos are required for visualization.