Currently, Patch inference using PatchInferer supports multi-threading via dataloader, which is fine for most of our use cases since IO is occurring at datatloader and splitter components of patch inference just do the slicing of the already loaded image.
However, in the case of whole slide images (WSI) IO (loading patches) should happen after it is split, which leads to the need for PatchInferer to take care of multi-thread calling to the splitter to create batch of patches. This may benefit the overall inference performance for WSI.
Currently, Patch inference using
PatchInferersupports multi-threading via dataloader, which is fine for most of our use cases since IO is occurring at datatloader and splitter components of patch inference just do the slicing of the already loaded image.However, in the case of whole slide images (WSI) IO (loading patches) should happen after it is split, which leads to the need for
PatchInfererto take care of multi-thread calling to the splitter to create batch of patches. This may benefit the overall inference performance for WSI.