Max Freudenberg
Max Freudenberg
The heatmap_interpolation.png reference image differs slightly, that's probably why the test failed. If you tell me where to find the code generating the reference output, I can try to fix...
Margins are now calculated per individual pixel, which should get rid of the difference in the test image.
Ok, I think it's not so easy to get this perfectly right, as padding the pixels is already a bit hacky. As the heatmaps only look wrong if NaN is...
https://github.com/MakieOrg/Makie.jl/blob/f92b2607de3788a6c2b45eb5fd5546fb634fab3a/CairoMakie/src/primitives.jl#L785-L795 here, right?
Just a side note: The pytorch timings are probably sub-optimal. One would use `torch.compile()` nowadays and `with torch.no_grad()` for inference, to actually suppress the computation of gradients. None that `torch.compile()`...
Disclaimer: I haven't worked on the NNlib codebase for quite a while, so the following assumes that there is no rotation code present as of now. Searching for "rotat" in...
4D usually corresponds to a batch of 2D images with a color channel. That is the most common use case for rotations. But if you want to implement 5D arrays,...
I think this can be closed, right?
Is there a way to force the complete summary printout?
Here are the timings on my machine (12 core cpu + 1080Ti). Setting `@testset verbose=true` did the trick :) ``` NNlib.jl | 164 164 2m30.9s CPU | 82 82 1m18.5s...