search

princeton-vl / DROID-SLAM

BSD 3-Clause "New" or "Revised" License
1.65k stars 272 forks source link

Question on correlation_kernels.cu #100

Open minooisbusy opened 1 year ago

minooisbusy commented 1 year ago

Thanks for the amazing research. I had a few questions, so I left questions.

The questions are:

  1. Is motion_filter.py:69 correct at the beginning of the correlation lookup code?
  2. I wonder why the range of i and j is defined up to rd+1 in the nested for loop of correlation_kernels.cu. This will make the neighborhood grid asymmetric. Any special reason?
  3. Since I am not familiar with CUDA, I am reading the code while trying to understand the flow of variables with my eyes. If I run the kernel for pyramid level 0 of the correlation volume in lines 52 to 65 of correlation_kernels.cu, dx=dy=0, so only the last if statement actually works as true?

Thanks in advance for your reply.