Official code for Understanding Bird’s-Eye View of RoadSemantics using an Onboard Camera - RAL/ICRA 2022

Link to paper

Check out our work on structured scene representation:

STSU (ICCV'21): https://github.com/ybarancan/STSU

TPLR (CVPR'22): https://github.com/ybarancan/TopologicalLaneGraph

Written with Tensorflow.

Make sure you have installed Nuscenes and/or Argoverse devkits and datasets installed

Data preparation

For Nuscenes, run make_nuscenes_labels.py and dataset_creator.py, by setting the necessary paths in the experiments/nuscenes_objects_base.py file. For Argoverse, run make_argoverse_labels.py, by setting the necessary paths in the experiments/argoverse_objects_exp.py file.

Training

For training, run the relavant dataset's train.py file.

Trained Models

We provide trained model for Nuscenes dataset:

https://data.vision.ee.ethz.ch/cany/BEV-stitch/bev-stitch-nusc.zip