batch_size and trainning time

[cv-lab-x]

hi, thanks for your great work!

1. why you use batch_size = 1 in your code, did your try to imcrease batch_size ? i got some errors when i just increaset batch_size from 1 to 2.
2. can this codes run with multi-gpu to improve training time?
3. what params can i try to improve the quality of the sdf and psnr?

looking forward to your reply! thanks! @niujinshuchong

[niujinshuchong]

Hi,

1. Because we need to align rendered depth with monocular depth so they should come from the same image. I think use more images at a time is also possible but you need to adapt the code to align them separately.
2. Yes, using multi-gpu will converge faster.
3. You can use grids representation to get better psnr and more detailed sdf but the sdf might be noise.