A question about the "__anyhit__ah" in "cuda_test/gaussians_aabb.cu".
Closed this issue · 2 comments
Hi,
Nice work, and thank you for sharing it!
I have a question about the implementation of "__anyhit__ah" in "cuda_test/gaussians_aabb.cu".
During the forward process in training, the function "computeBufferForward" is used to calculate the "density_buffer" of each slab_space, which is easy to understand. However, during testing, it seems that this calculation is performed in "__anyhit__ah", and I found that the "r_2" value here is identical for every sample point in the same slab_space, which means the weights for each sample point are the same. Did I understand it correctly?
And I wonder why it differs from the forward process during training.
Very appreciate it if provide any clues.
Apologies, I missed the update of M_xhit_xgaus at the end of the for-loop. That clarifies my question.
Thank you again.
Thank you for your interest in our work. Indeed, M_xhit_xgaus is updated at the end of the for-loop with: M_xhit_xgaus += M_d_dt;.