Open yipingp opened 12 months ago
Thank you for being interested in our work. I'm glad to hear that you've noticed an improvement in mesh quality using our method. The decrease in speed is expected due to spheres' optimization, but it should not be large (less than 15% slower per iteration).
Thanks for your reply. get_random_sphere_rays_at()
samples rays based on sphere origins, which varies through iterations. I wonder is there any chance to speed up with DDP?
Thanks for your great work. I trained neus with sphere guided, the mesh quality improved significantly. However, I've noticed a decrease in speed once the iteration reaches 5000 (i_warmup_spheres). Is there any way to increase the speed?