You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This is just to let you know that recently we added code for nearest neighbour and bilinear upsampling to NNlib here in these PRs 1, 2. The code uses WHCN memory layout (and works on the first two dimensions). The nearest version is single threaded for CPU and works out of the box for GPU. The bilinear version is multi-threaded for CPU and has a GPU implementation here. Note that bilinear upsampling aligns the corners, so I think it's a bit different from what you are doing, but I intend to make the behaviour optional to be able to match your implementation. Both versions are "super fast" but I haven't benchmarked against ImageTransformation.jl. RGB data is currently not supported out of the box, but writing a wrapper should be simple.
Hi everybody!
This is just to let you know that recently we added code for nearest neighbour and bilinear upsampling to NNlib here in these PRs 1, 2. The code uses WHCN memory layout (and works on the first two dimensions). The nearest version is single threaded for CPU and works out of the box for GPU. The bilinear version is multi-threaded for CPU and has a GPU implementation here. Note that bilinear upsampling aligns the corners, so I think it's a bit different from what you are doing, but I intend to make the behaviour optional to be able to match your implementation. Both versions are "super fast" but I haven't benchmarked against ImageTransformation.jl. RGB data is currently not supported out of the box, but writing a wrapper should be simple.
See also JuliaMath/Interpolations.jl#396 - would be cool to recycle code where possible, perhaps some day in the future :)
The text was updated successfully, but these errors were encountered: