Prioritize using imagesize library to get image size for ImageFromFile

[imyhxy]

Summary

Accelerate loading of image file-based datasets.

I found that printing out the YOLO dataset information for the first time was slow. After some digging I found that datamaro was reading the entire dataset through to get the size of each image.

ds = Dataset.import_from("/yolo-ultralytics", "yolo")
print(ds)  # <-- wait a long time

    # from class Image
    @property
    def size(self) -> Optional[Tuple[int, int]]:
        """Returns (H, W)"""

        if self._size is None:
            try:
                data = self.data  # <-- load the whole media into memory
            except _image_loading_errors:
                return None
            if data is not None:
                self._size = tuple(map(int, data.shape[:2]))
        return self._size

Interactive encoding with datasets on HDD is slow. So I added an override size() property in the ImageFromFile class which first tries to get the image size using PIL. The PIL library is about 8 times faster than OpenCV in getting the image size.

All dataset classes that use the size property of ImageFromFile can benefit from this modification.

How to test

Checklist

• I have added unit tests to cover my changes.​
• I have added integration tests to cover my changes.​
• I have added the description of my changes into CHANGELOG.​
• I have updated the documentation accordingly

License

• I submit my code changes under the same MIT License that covers the project.
  Feel free to contact the maintainers if that's a concern.
• I have updated the license header for each file (see an example below).

# Copyright (C) 2023 Intel Corporation
#
# SPDX-License-Identifier: MIT

[imyhxy]

@vinnamkim
It would be great if adding new dependencies was acceptable. I actually considered several approaches here and settled on PIL for compatibility. But yes, imagesize is a little faster.

imagesize vs cv2: 0.3s vs 8.1s on my benchmark dataset.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (784e039) 80.54% compared to head (38caf4d) 80.51%.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #1259      +/-   ##
===========================================
- Coverage    80.54%   80.51%   -0.04%     
===========================================
  Files          271      270       -1     
  Lines        30438    30426      -12     
  Branches      5930     5931       +1     
===========================================
- Hits         24517    24498      -19     
- Misses        4532     4535       +3     
- Partials      1389     1393       +4     

Flag	Coverage Δ
ubuntu-20.04_Python-3.8	?
windows-2022_Python-3.8	80.51% <100.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[vinnamkim]

LGTM