Undesirable result on example image
Hi, thanks for the great work.
I have trained the model with the example panorama image kitchen you provided.
However the result seems pretty bad.
Looks like the depth estimation or the inpainting model failed to some extend.
Do you have any clue of this issue?
Input Image
Reconstruction Result
https://github.com/perf-project/PeRF/assets/59246000/aa315054-89b9-4638-9ee8-632652f5cf16
Most part of the pointcloud reconstructed from the monocular depth estimator looks good to me. However, I think it might be challenge for the inpainting model to restore the occluded area behind the kitchen island?
BTW, The fridge looks normal in the pointcloud but why it seems to be distorted in reconstruction views?
How did I get the pointcloud, was it generated by PeRF or did you process it yourself, I need the pointcloud