Q: Pixel-Perfect-SfM in a simulated environment

[TommasoLabieni]

Hello and thanks a lot for this incredible project.

I successfully tested this SfM framework using public datasets achieving stunning results. Now, before creating my custom dataset, I would like to test pixel-perfect-sfm more in a simulated environment such as CARLA. The core idea is to have an image dataset taken from this simulated environment and having the possibility to reconstruct the 3D scene using several images grabbed from simulated camera sensor(s) and evaluating the results using ground-truth data coming from simulated LiDAR sensor(s).

This would be extremely relevant for my application since it reflects exactly my real-world setup.

My questions are:

1. Do you think this is feasible?
2. Is there any best-practice I show consider when generating the image dataset?

Thank you!