malteprinzler / diner

Official PyTorch Implementation of the CVPR23 Paper "DINER: Depth-aware Image-based NEural Radiance fields"
79 stars 8 forks source link

Question about the standard deviation #5

Open Sibo2rr opened 10 months ago

Sibo2rr commented 10 months ago

Hi! Thank you for your amazing work!

I have a question about fitting a polynomial of 1st order for confidence map and standard deviation. In another issue you mentioned analyzing the relation between the predicted confidence and the deviation between prediction and ground truth depth value, can you explain it in detail? Since I noticed that the output of TransMVSNet ranges from 0 to 1 while the ground truth depth of DTU ranges from 0 to 1000, how did you align it? Besides, if I understand correctly, did you calculate the standard deviation between the produced depth map and truth depth value for each pixel (like use (D_ij-D_GT_ij)^2/#pixel as deviation), and then find a polynomial of 1st order to represent confidence into the standard deviation of each pixel?

Thanks in advance.