UrbanLaneGraph Dataset & Benchmark

The UrbanLaneGraph dataset is a first-of-its-kind dataset for large-scale lane graph estimation from aerial images.

---

Website

---

UrbanLaneGraph Dataset

To view the download instructions, please see the file DOWNLOAD.md

Dataset processing

Instructions on how to process the dataset into a representation suitable for model training may be found in dataset_preparation/PROCESS_DATA.md.

Model training

Based on a number of requests regarding pre-trained model checkpoints, we provide them here.

The models defined in methods/ can be trained using the scripts methods/train_centerline_regression.py and methods/train_lanegnn.py. For more information, please see methods/TRAINING.md.

Evaluator

To install the urbanlanegraph_evaluator package, find the instructions in urbanlanegraph_evaluator/INSTALL.md.

Instructions on how to evaluate your model locally using the package may be found in urbanlanegraph_evaluator/EVALUATION_INSTRUCTIONS.md.

---

UrbanLaneGraph Benchmark on Eval.AI

Coming soon!

Acknowledgements

We thank the following projects for making their code available:

• LaneExtraction
• apls