Closed jere357 closed 6 months ago
It's not redundant, if size of Resize is a scalar it resizes the smallest side to that value, so it's not square. Crop is needed.
yeah, i misread the torch.Resize docs superbad + i was working with only square images, then had another redundant idea because the code was working as it should, keep up the excellent work!
Code from readme.md:
This preprocess object contains 5 augmentations.
I think that the CenterCrop call is redundant since the image is resized to 224 in the step before, why call CenterCrop(size=(224,224)) to an image that has been resized to 224x224 ?
The preprocess object gets constructed in the image_transform() function call in transform.py (around line 372) Default values are resize_mode = 'shortest'
Discussion welcome, I'd be down to make a PR fixing this if you decide it's worth fixing