MotionCNN

This repository contains an CNN-based model used in Muñoz Sánchez et al. "Robustness Benchmark of Road User Trajectory Prediction Models for Automated Driving" presented at The 22nd World Congress of the International Federation of Automatic Control (IFAC 2023). This is a forked and adapted repository of MotionCNN. The original README is below.

---

Waymo motion prediction challenge 2021: 3rd place solution

• 📜Paper

• 🗨️Presentation

• 🎉Announcement

• 🛆Motion Prediction Channel Website

• 🛆CVPR2021 workshop

• UPDATE❗ Related repo with 3rd place solution code for Waymo Motion Prediction Challenge 2022

Team behind this solution:

1. Artsiom Sanakoyeu [Homepage] [Twitter] [Telegram Channel] [LinkedIn]
2. Stepan Konev [LinkedIn]
3. Kirill Brodt [GitHub]

Dataset

Download datasets uncompressed/tf_example/{training,validation,testing}

Prerender

Change paths to input dataset and output folders

python prerender.py \
    --data /home/data/waymo/training \
    --out ./train
    
python prerender.py \
    --data /home/data/waymo/validation \
    --out ./dev \
    --use-vectorize \
    --n-shards 1
    
python prerender.py \
    --data /home/data/waymo/testing \
    --out ./test \
    --use-vectorize \
    --n-shards 1

Training

MODEL_NAME=xception71
python train.py \
    --train-data ./train \
    --dev-data ./dev \
    --save ./${MODEL_NAME} \
    --model ${MODEL_NAME} \
    --img-res 224 \
    --in-channels 25 \
    --time-limit 80 \
    --n-traj 6 \
    --lr 0.001 \
    --batch-size 48 \
    --n-epochs 120

Submit

python submit.py \
    --test-data ./test/ \
    --model-path ${MODEL_PATH_TO_JIT} \
    --save ${SAVE}

Visualize predictions

python visualize.py \
    --model ${MODEL_PATH_TO_JIT} \
    --data ${DATA_PATH} \
    --save ./viz

Citation

If you find our work useful, please cite it as:

@article{konev2021motioncnn,
  title={MotionCNN: A Strong Baseline for Motion Prediction in Autonomous Driving},
  author={Konev, Stepan and Brodt, Kirill and Sanakoyeu, Artsiom},
  year={2021}
  booktitle={Workshop on Autonomous Driving, CVPR}
}

Related repos

• Kaggle Lyft motion prediciton 3rd place solution