CielAl

I am still reading the codes. Actually lots of the protected methods like `model.predict` already takes care of the case of dataloader input, and mostly it is just the highest...

> Yup, this looks to be inline with what i was thinking, thanks! top priority is the dicom component (using wsidicom), and then we can regroup on the other components...

> Openslide has now incorporated DICOM support: https://openslide.org/news/. > > Let's keep an eye on this. The latest openslide version is currently available via ppa repository only. Once the ubuntu...

> Hi everyone, > > while I was trying to use HistoQC on DICOM files, I stumbled across this Issue and was wondering what the current status is. > >...


OK it seems like my input of fN was arbitrary, hence the dimensions did not agree. That won`t be an issue if all functions are correctly involked.

I believe the out-dim is just the dimension of the feature representation produced by the projection head. The reason the authors use out_dim in the models.resnet18(pretrained=False, num_classes=out_dim) is only because...

PS atm the changes were pushed into dicom_support branch as the wsihandles are not merged into the main. Refer to #288

Seems like a simple regionprop + minimum area thresholding can work. Just return the length of `skimage.measure.regionprops` output list and that's it.

Question would be - should that be in the final processing stage, or that the # of tissue pieces should be traced at the end of each module (as post...