Add Open Images format

[sizov-kirill]

Motivation and context

How has this been tested?

Checklist

• I submit my changes into the develop branch
• I have added description of my changes into CHANGELOG file
• I have updated the documentation accordingly
• I have added tests to cover my changes
• I have linked related issues (read github docs)
• I have increased versions of npm packages if it is necessary (cvat-canvas,
  cvat-core, cvat-data and cvat-ui)

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) 2021 Intel Corporation
#
# SPDX-License-Identifier: MIT

[zhiltsov-max]

Review the implementation wrt #3790

[zhiltsov-max]

Maybe, we should be more specific about the version and add v6.

[sizov-kirill]

Done.

[nmanovic]

@kirill-sizov , image meta information is provided only for annotatated images. Is it expected?
I have approximately 200 images in my task. I have annotated 9 first frames.

$ cat images.meta
# <image name> <height> <width>

Abyssinian_123 333 500
Abyssinian_71 539 833
Bengal_197 206 300
Birman_128 375 500
Birman_43 500 340
Birman_44 375 500
Birman_82 324 448
Bombay_104 250 167
Bombay_157 849 704

[sizov-kirill]

@kirill-sizov , image meta information is provided only for annotatated images. Is it expected? I have approximately 200 images in my task. I have annotated 9 first frames.

$ cat images.meta
# <image name> <height> <width>

Abyssinian_123 333 500
Abyssinian_71 539 833
Bengal_197 206 300
Birman_128 375 500
Birman_43 500 340
Birman_44 375 500
Birman_82 324 448
Bombay_104 250 167
Bombay_157 849 704

@nmanovic, Yes it's expected, because we need information about image size only for images that have annotation objects.

[nmanovic]

@kirill-sizov , image meta information is provided only for annotatated images. Is it expected? I have approximately 200 images in my task. I have annotated 9 first frames.

$ cat images.meta
# <image name> <height> <width>

Abyssinian_123 333 500
Abyssinian_71 539 833
Bengal_197 206 300
Birman_128 375 500
Birman_43 500 340
Birman_44 375 500
Birman_82 324 448
Bombay_104 250 167
Bombay_157 849 704

@nmanovic, Yes it's expected, because we need information about image size only for images that have annotation objects.

I'm not sure I understand "we". Cityscapes dataset can be used by multiple scripts around the world which can rely on the file. What is the answer from the dataset format point of view?

[sizov-kirill]

@kirill-sizov , image meta information is provided only for annotatated images. Is it expected? I have approximately 200 images in my task. I have annotated 9 first frames.

$ cat images.meta
# <image name> <height> <width>

Abyssinian_123 333 500
Abyssinian_71 539 833
Bengal_197 206 300
Birman_128 375 500
Birman_43 500 340
Birman_44 375 500
Birman_82 324 448
Bombay_104 250 167
Bombay_157 849 704

@nmanovic, Yes it's expected, because we need information about image size only for images that have annotation objects.

I'm not sure I understand "we". Cityscapes dataset can be used by multiple scripts around the world which can rely on the file. What is the answer from the dataset format point of view?

The file images.meta is generated by Datumaro to read the dataset more efficiently. The format stores annotations in a normalized form relative to the size of the image. To avoid reading images, Datumaro uses a workaround in the form of this meta file. Original dataset doesn't contain such information.