Results from MipNeRF-360 room scene

[simonakocour]

Hi, thank you for this work!
I am trying to replicate your results for room scene in MipNeRF 360 dataset. But removing brown shoes on the ground does not give me good results, compared to yours in the paper. I tried using different views to fine-tune the in-painting and used the new depth map as input for running inference in-painting. But it did not help.

My masks are from Grounded SAM and Infusion checkpoints from HuggingFace.

Could you share any tips, how to reach similar results to yours? Thank you!

[Johanan528]

Thanks for your interest. I think there might be two suggestions :

1. You could try dilating the masks when removing the shoes and remove outliers after removing the Gaussians of the shoes.
2. Could you visualize the inpainted Gaussians?