About BEV Encoder

[Jeong-Bin]

Hello, thank you very much for sharing your code.

I'm going to do a job of transforming the BEV ground truth map into images.
Therefore, I thought your work(specifically BEV Encoder) would be helpful and referred to stag_1_bev.yaml.

However, in the vqgan and VQSegmentationModel, the code of CVT style that transforms the view from images to BEV map is checked in VQModel's encoder, but I did not find any particular view-transformation code from BEV map to images.
(Of course, it could be my mistake.)
If there is a code related to this, please let me know which part it is specifically.

Thank you.

[alexanderswerdlow]

Hi! Just to be clear, are you asking for code that goes from BEV map -> First-person view [e.g., car cameras] with the segmentation of objects/map [see the image below, taken from BEVControl]?

If so, I don't think that function is in this repo. I had it at one point but I don't think it's available here. Luckily, I think several other repos have something similar if you wanted that, but I didn't use it for any of the results in the paper.

[Jeong-Bin]

Thank you for your kind reply!
What you understand is correct.
Unfortunately, the "BEV map -> first-person view" process is not here.
However, thank you for responding in detail with additional materials.

[Jeong-Bin]

Oh, I just saw that you have a repo similar to what I want.
Could you please let me know the relevant link?
+The [BEV map -> Multi cameras view] don't have to be with the segmentation of objects/map.

[alexanderswerdlow]

Hey! Unfortunately, I wasn't able to find my code, but you could use something like this for argoverse 2: https://github.com/argoverse/argoverse-api/blob/master/demo_usage/argoverse_map_tutorial.ipynb