Skip to content

Latest commit

 

History

History
38 lines (17 loc) · 1.3 KB

PROCESS_DATA.md

File metadata and controls

38 lines (17 loc) · 1.3 KB
Raw

Dataset processing for model training

In this document, we describe how to process the UrbanLaneGraph dataset for model training.

0. Download the dataset

For instructions on how to download the dataset, please refer to urbanlanegraph_dataset/DOWNLOAD.md.

1. Generate raw samples (for regressor model traning)

python generate_raw_samples.py /path/to/urbanlanegraph/dataset/ /path/to/raw/output <city_name> <split>

The parameter <city_name> can be either of miami, paloalto, pittsburgh, austin, washington, detroit. The parameter <split> can be either of train, val.

We do not provide the samples for the dataset test split.

Make sure that the directory lanegnn/dataset_preparation is in your PYTHONPATH environment variable. (To do so, run export PYTHONPATH=$PYTHONPATH:/path/to/lanegnn/dataset_preparation)

2. Generate pth samples (for LaneGNN model training)

python generate_pth_samples.py --config ../methods/lanegnn/config/config.yaml --raw_dataset /path/to/raw/output/ --processed_dataset /path/to/processed/output/ --ego_regressor_ckpt /path/to/checkpoint.pth --context_regressor_ckpt /path/to/checkpoint.pth

Note that you have to run the script for all the directories you might have created in step 1.