How should we correct the inference with mirrors in the input?

#9
by bensaini7 - opened

For example, if a door has mirrors on it, then the model's prediction is as if the door's body is hollow with depth volume.
This is similar to the red-eyes problem, which we used to see back in the day with gen1 digi cameras. I am confident a fix should be imminent.

Screenshot from 2025-01-21 09-09-54.png

Sign up or log in to comment