search

David0tt / DeepGTAV

A system to easily extract ground truth training data for different machine learning tasks from GTAV
GNU General Public License v3.0
89 stars 9 forks source link

road semgentation and multiple side-view cameras #9

Closed eliabruni closed 2 years ago

eliabruni commented 2 years ago

Hello,

Great work!

I have two questions:

  1. Do you think it would be possible to obtain also road segmentation (e.g., drivable/non drivable)?
  2. Do you think it would be possible to mount multiple side-view cameras on top of a car while also getting the aerial view?

Thanks!

David0tt commented 2 years ago

Hey,

  1. I don't know of an easy way that this would be possible. In general the segmentation information is obtained from the stencil buffer of the GTAV rendering pipline. As I recall the roads don't have an individual stencil mask, so this would not be easy. For more information you can look into https://www.adriancourreges.com/blog/2015/11/02/gta-v-graphics-study/

  2. Yes it would be possible to get additional camera views. To do this one would need to add additional cameras in the DataExport.cpp and while having the game paused for the capture one would need to switch to the different active cameras, capture the image + data from their view, then switch to the next, capture from their view and so on. Finally one would transfer all the data by adding it to the message. This is possible, however it is quite some coding work.

If you actually implement this, a pull request would be much appreciated :D

eliabruni commented 2 years ago

Hello @Eisbaer8 , many thanks for the clarifications! I will look into it and, of course, I'll do a PR if I manage to implement it.

Cheers

GANWANSHUI commented 1 year ago

Hi, @eliabruni. I also have an interest in seeing the second question you mentioned. Was there any progress from your side? Thanks a lot!

Do you think it would be possible to mount multiple side-view cameras on top of a car while also getting the aerial view?

eliabruni commented 1 year ago

Hey @GANWANSHUI , not really, unfortunately. It seems quite a task the way I see it.