Segmentation dataset datablock api

Part 1 (2019)
1

Still learning how to use the data block API, and I have a segmentation dataset that is formatted like this:

images/
–train/<long_integer>.jpg
–val/<long_integer>.jpg
annotations/
–train/<matching_integer>.png
–val/<matching_integer>.png

Would anyone please help me out in setting up the data block api?

I could also modify the structure to this if it’s more straightforward:

train/
–images/<long_integer>.jpg
–annotations/<matching_integer>.png
valid/
–images/<long_integer>.jpg
–annotations/<matching_integer>.png

I think if I do it the second way I can start out like this:

data = (SegmentationItemList.from_folder(path)
        .split_by_folder()

But I’m getting hung up on how to set the labels.

2

Your first folder structure works. You just have to give as path the path where the images are and define an appropriate function that transforms your image path to a mask path.

Hence, your path is: root_path/‘images’