Closed Selur closed 2 years ago
VSGAN already supports this via the chunk
parameter of VSGAN.run()
, however, I assume you made this because of the artifacts around the edges and are looking for a solution.
I saw your posts on doom9 about this, and saying a possible solution would be to pad the images before running the model and then cropping it off before re-merging the images. This is definitely a possible solution and I will test this in the future.
Yeah, thanks for looking into it, also some control over the size of the 'chunking' would be nice.
Would be nice if VSGAN would some sort of tiling support, to lower the memory requirement. Tiling as in splitting the source image into smaller images apply the model and stick the processed images back together.