Open ajinkya-kulkarni opened 1 year ago
Our preferred method of adding additional algorithms is to wrap them in a new docker image and expose the parameters via the slicer cli execution model. This repo can certainly be used as a starting point for that, but a new docker can often be simpler with less dependencies.
Hello, I have a question regarding how to build my own analysis using models developed with PyTorch? (eg. patch based classification)
@AhmedElshahedd There are so many approaches that the best one depends on what you are trying to accomplish. For instance, if you are trying to analyze whole slide images at full resolution a patch at a time, the methods used in HistomicsTK are appropriate (a docker with torch installed and using a tile iterator to process each patch). If you want to only process user selected patches quasi realtime, Monai Label or https://github.com/DigitalSlideArchive/dsa-run-custom-ai-models would be more appropriate. See also https://github.com/DigitalSlideArchive/superpixel-classification for a way that works on multiple images and uses tensorflow (but there is a branch that is in progress that uses torch).
Hi Devs, thanks for this repo! Are contributions to the nuclei segmentation part using an external library allowed? For example using CellPose or something similar.