openmedlab/BROW

BROW/Slide-level multi-class subtyping task /feature_extract----error_code???

Closed this issue · 3 comments

464hee commented

image
I keep a question about the image processing pipeline of extract_features.py, why is it necessary to use transforms.CenterCrop((224,224)) instead of transforms.Resize((224,224)) in the feature extraction, @wustone1995

464hee commented

415DCFD0-9B38-464b-A86A-06BF38C0D1B7
transforms.CenterCrop((224,224))Crop the 224224 size from the center region, if you accidentally input an image larger than 224224, then you can't get the full image features?

In the slide-level-subtyping task, we employed a multiple instance learning framework, which necessitates the cropping of the original WSI images. The patch size is approximately 224, where the information acquired through center-cropping and resizing remains similar. This is one of the approaches adopted in the experiments.

464hee commented

ok