jzuern / lanegnn

[CVPR 2023] Repository for the UrbanLaneGraph Dataset & Benchmark and the LaneGNN approach

Hi there 👋

lanegnn's Issues

Need to obtain translation matrix between ULG aerial image and Av2 city coordinate system?

Hello,

I am trying to work with satellite images in UrbanLaneGraph Dataset, great works.
I need to be able to translate coordinates between the satellite image coordinate system and the Av2 city coordinate system around all 6 city.

Can you please provide any information or resources that would help me obtain this correspondence? I appreciate any assistance you can provide.

Thank you.

Training guidance

Hi, thank you for your excellent open-source work!

I would like to train the model according to the code, but I may encounter some errors such as missing data preparation modules. Could you please provide some guidance or readme file for training the model?

Thanks for your reply.

Inference Guidance and Context Images Clarification

Hi,

Thank you for open-sourcing this work.

I'm trying to run inference on a model I have trained for the Successor-LGP task; specifically I would like to evaluate on the files you share in the dataset in the urbanlanegraph-dataset-pub-v1.1/{city}/successor-lgp/eval/* directories.

In this eval directory, there exist *-graph.gpickle, -rgb.png and -viz.png files. Please could you point me in the direction of any files or functions I could use to run inference on these, in order to generate the pkl files required to run evaluation?

The reason for my confusion is that the rest of the codebase appears to use additional *-rgb-context.png images, which are of size 512x512, whereas the images included in the Successor-LGP eval directory are all size 256x256, and these context images do not exist. As far as I can tell, the Urban Lane Graph paper also does not reference use of these larger context images, but please correct me if I am wrong. Are these context images required, and if not, how can I run inference (and evaluation) without them?

Thank you in advance.

Pretrained model

Hello,
I want to use your code and dataset for better understanding of your great work in CVPR.

Will a pretrained model released later? Cause i have no enough GPU for training a good model.

Thanks for your reply.

About AutoGraph

Hi Dr. Jannik Zürn,

I am interested in another related work of yours - AutoGraph. AutoGraph adds trajectory data of different agents on top of the UrbanLaneGraph dataset. Although the paper provides the UrbanTracklet dataset, it seems to be raw.

I would like to know more about the trajectory segmentation method in the UrbanTracklet dataset and how it aligns with the UrbanLaneGraph dataset. Please tell me more details about processing the UrbanTracklet dataset.

Thank you in advance for your time and assistance!

Best regards,
HengQi

requirements.txt

Hi, is the requirements.txt correct? It seems not to be in normal format, and i got an error when installing:

> pip install -r requirements.txt
ERROR: Invalid requirement: '_libgcc_mutex=0.1=conda_forge' (from line 4 of requirements.txt)
Hint: = is not a valid operator. Did you mean == ?

ModuleNotFoundError: No module named 'dataset_preparation'

Hi, I tried to run this code but got error? can you also opensource this dataset_preparation module

Inquiry about Creating LaneGNN Datasets using QGIS