Performance in low light environment

I have noticed that my model runs into problems when it comes to low light environments. Part of the problem is my camera, but I'm asking if anyone has dealt with a similar problem and what solution did you find, if any. I have noticed that in a dimly lit room the depth estimation becomes worse. If a part of the room is dimly lit, I can see the depth predictions scale to be way worse than in the other corner with better lighting. For example, the drawer looks like a small bump instead of a substantial bump in formation of the wall, but if I have better lighting the prediction becomes drastically better.
I'm currently running the pipeline on an embedded device so I can't run the bigger models. Do they have the same issues?
And if anyone has dealt with the same low light problem, what did you do to improve it besides turning on the lights?
The culprit is most likely the training data—it lacks the low light examples. When the light level drops so low that we start getting image noise, the prediction becomes completely useless; it just becomes flat. But that is not the main concern I have. The main concern is that even not so substantial changes in lighting change the prediction accuracy substantially