Awesome

Waymo motion prediction challenge 2022: 3rd place solution (May, 26)

Our implementation of MultiPath++

General Info:

Team behind this solution:

Stepan Konev

Code Usage:

First we need to prepare data for training. The prerender script will convert the original data format into set of .npz files each containing the data for a single target agent. From code folder run

python3 prerender/prerender.py \
    --data-path /path/to/original/data \
    --output-path /output/path/to/prerendered/data \
    --n-jobs 24 \
    --n-shards 1 \
    --shard-id 0 \
    --config configs/prerender.yaml

Rendering is a memory consuming procedure so you may want to use n-shards > 1 and running the script a few times using consecutive shard-id values

Once we have our data prepared we can run the training.

python3 train.py configs/final_RoP_Cov_Single.yaml

If you find this work interesting please ⭐️star and share this repo.

Citation

If you find this work useful please cite us

@misc{https://doi.org/10.48550/arxiv.2206.10041,
  doi = {10.48550/ARXIV.2206.10041},
  url = {https://arxiv.org/abs/2206.10041},
  author = {Konev, Stepan},
  keywords = {Computer Vision and Pattern Recognition (cs.CV), FOS: Computer and information sciences, FOS: Computer and information sciences},
  title = {MPA: MultiPath++ Based Architecture for Motion Prediction},
  publisher = {arXiv},
  year = {2022},
  copyright = {arXiv.org perpetual, non-exclusive license}
}