Thomas Müller
Thomas Müller
Hi, the slight difference to the paper is intentional: the number of vertices in the code is `N_i` rather than `N_i+1` such that the lower levels of dense grids are...
Ah, that's a good point -- the 0.5-offset does indeed cause wrap-around (and thus a weird type of collision) in the dense lower-level grids. I haven't experienced any issues caused...
It does -- although it's interesting that disabled VSync would cause flicker artifacts beyond tearing. Thanks for the fix!
Fixed by https://github.com/mitsuba-renderer/nanogui/issues/102#issuecomment-1287312558
Fixed on latest `master`
Hi, I've looked into creating a custom TensorFlow op in the past, but it seemed very cumbersome. Apparently requires a bazel TensorFlow build system on Linux, and I can't find...
Yes, that's correct! The command line was ```sh tiny-cuda-nn> .\build\bench_image_ours.exe .\data\images\albert.exr .\data\config.json ``` with `n_neurons: 128` and `n_neurons: 64`, respectively. The benchmark was run on Windows / MSVC 2019 /...
> In any case, I observe lower performance than yours, except for the case of neurons=128, where I get 2x throughput, which is actually faster than the case of neurons=64...
> Also, what is the extent of the modifications of CUTLASS wrt the latest version available on github? I saw the PreReLU options in GemmShape, but those are only used...
Hi there! 1. It's a continuum, which I would classify as follows: a. Let's use CUTLASS's matrix multiplication routines (implemented in `CutlassMLP`) as a baseline, since these avoid unrelated overheads...