SegmentationItemList

fastai
1

Hi
I would like to update my project from fastai v1 to v2 and especially to add the timm library.
I slowly work from cell to cell, I’m a low level hobbyist, and now I’m stuck with this:

images_with_macula = SegmentationItemList.from_folder(‘anatomie_png/images’).filter_by_func(lambda path: not (‘nomacula’ in str(path))).split_by_folder()

What’s the replacement for SegmentationItemList in v2?

Thank you for your help.
Christian

2

SegmentationDataLoaders should be a replacement for SegmentationItemList.

The computer vision tutorial in the documentation also covers semantic segmentation: fastai - Computer vision intro

1 Like
3

Hi Marius
Thank you, I will have a look into it.

4

Hi Marius

I really stuck here how to change the code from the following section to Version 2, maybe you could help me to get this thing work?

Original code:

images_with_macula = SegmentationItemList.from_folder('anatomie_png/images').filter_by_func(lambda path: not ('nomacula' in str(path))).split_by_folder()

from pathlib import PosixPath

def get_label(path_to_image):
  return PosixPath('segmentations_png/segmentations')/path_to_image.parts[2]/path_to_image.parts[3]/path_to_image.parts[4]

codes = ['nomacula','macula']

labelled_images_with_macula = images_with_macula.label_from_func(get_label, classes=codes)

data = labelled_images_with_macula.transform(get_transforms(), size=224, resize_method = ResizeMethod.PAD, padding_mode='zeros', tfm_y=True).databunch(bs=4)

My try:

#path= get_files.from_folder('anatomie_png/images').filter_by_func(lambda path: not ('nomacula' in str(path))).split_by_folder()

from pathlib import PosixPath

def get_label(path_to_image):
  return PosixPath('segmentations_png/segmentations')/path_to_image.parts[2]/path_to_image.parts[3]/path_to_image.parts[4]

fnames = get_image_files(path)

codes = ['nomacula','macula']

data = SegmentationDataLoaders.from_label_func(path, fnames, label_func=get_label, codes=codes)

Thank you!

5

Sorry, actually I don’t really understand your code example. What error do you get? do you get any error?

How exactly is your data structured?

Could you maybe give a more throughout example?

Did you read the tutorial on semantic segmentation in the docs? fastai - Computer vision intro

1 Like