search

princeton-vl / RAFT-Stereo

MIT License
721 stars 139 forks source link

camera noise #61

Closed gpuartifact closed 2 years ago

gpuartifact commented 2 years ago

Hello, Would you recommend to either: a) add the expected camera noise to the simulation data for training to implicitly train de-noise or b) should the camera image be de-noised first as good as it can be before passing it to RAFT-stereo?

I am trying to train a for a specific camera sensor.

lahavlipson commented 2 years ago

This sounds like an empirical question. I suggest try different options and pick the model with the best validation performance.