The 3rd place solution for competition “Lyft Motion Prediction for Autonomous Vehicles” at Kaggle

▶️
Video: link

Blogpost: link

Brief solution writeup: link

How to reproduce results

[Optional] Set the paths in the configs. But the default paths should work as well.

Set path where to store prerendered dataset in src/1st_level/config.py
Set path where the predicts of the 1st level models are saved in src/2nd_level/config.py

pip install -r requirements.txt
Apply patch to l5kit with ./apply_l5kit_patch.sh (it disables processing of rasterized images to allow rasterizer to return multiple results).

bash prepare_data_train.sh

bash train.sh

Run inference of 1st level models on the test set.
You may need to change which chekpoints to load when predicting (in predict_test_l1.sh), as the best epoch may change after you retrain the models.

bash prepare_data_test.sh
bash predict_test_l1.sh

cd src/2nd_level && python train.py

Make sure you’ve set all paths right in 2nd_level/config.py w.r.t. the 2nd_level directory.

cd src/2nd_level && python infer.py

The file witn final predictions will be saved to `src/2nd_level/submission.csv’.

Directory structure example (i.e., how it should look like after everything is trained and predicted) is in directory_structure.txt.

To skip training the 1st level models, you can download the pretrained weights by running bash download_1st_level_weights.sh.
To skip training and inference of the 1st level models, you can download all predicts. More details on this are in src/1st_level/submissions.
More details on how to use 2nd level model are in src/2nd_level.
Our final 2nd level model with 9.404 Private LB score is already committed in this repository (src/2nd_level/transformer.bin). To run inference using this model you can directly execute cd src/2nd_level && python infer.py.