- Simple and Lightweight: Requires no additional CUDA libraries, ensuring simplicity and ease of use.
- Efficient on Embedded Devices: Designed to be friendly and effective on embedded devices.
Masked road model
Masked trajectory model
Trajectory prediction
pip install -r requirements.txt# Prepare data
# Download from: https://www.argoverse.org/av2.html#download-link
data_root
├── train
│ ├── 0000b0f9-99f9-4a1f-a231-5be9e4c523f7
│ ├── 0000b6ab-e100-4f6b-aee8-b520b57c0530
│ ├── ...
├── val
│ ├── 00010486-9a07-48ae-b493-cf4545855937
│ ├── 00062a32-8d6d-4449-9948-6fedac67bfcd
│ ├── ...
├── test
│ ├── 0000b329-f890-4c2b-93f2-7e2413d4ca5b
│ ├── 0008c251-e9b0-4708-b762-b15cb6effc27
│ ├── ...# Set data_root in conf/config.yaml
python preprocess.py# Train mrm and mtm models
python train.py model=model_mrm epochs=20
python train.py model=model_mtm epochs=20
# Set mrm_checkpoint, mtm_checkpoint in conf/config.yaml for combined training
python train.py monitor=val_AvgMinFDE# Set checkpoint in conf/config.yaml for evaluation
python eval.py test=true