Tennis table dataset

[dfbakin]

Main changes

• Uploaded data to DVC
• Added parser from binary data (.out files) to single-channel .png
• added YAML file with intrinsic parameters and results of stereo calibration (translation and rotation vectors)

[dpaleyev]

Added final version of annotation. 4050 images from different sets were annotated.

Answering @BakinDF question, I'd prefer to left all images and annotation we have in master as it is now.

[AndBondStyle]

LGTM overall, haven't found anything critical. However, some things to consider:

1. Is there any raw video files? Looks like there are only images in dvc. Arguably, I'll store all videos in the raw format, and delegate any other manipulations (e.g. splitting the raw video into .png frames and converting bayer to rgb) to other scripts and pipelines. For labelling datasets, it should be possible to reference the raw frame by index inside the original raw video. But maybe it's too complicated, and just for labelling, data redundancy is acceptable.
2. It would be convenient to have some sort of description of each dataset. This includes: some sort of overview, or even a short preview video; which cameras were used and with what settings; how cameras was positioned relative to the table and each other (measured in real world); how the dataset is structured, how images/video is stored, how to work with it (e.g. example script to parse the images)
3. Folder names should be named consistently, so underscore_case or kebab-case is preferred
4. requirements.txt is still used, however there's also a modern pyproject.toml file in the project. Maybe we should consider moving requirements under poetry. We can revisit this when (if?) the @dpaleyev's ML training/inference pipeline will be added to this repo
5. binary_parser.py: useful tool, but I think we'll be extending its functionality in the future. Also, ffmpeg can be used to split the raw video into separate images, and it may be faster or more convenient in some cases. But having a python wrapper with a nice CLI is always ok