Add conversion from ethology dataset to torch dataset / dataloader

[sfmig]

Is your feature request related to a problem? Please describe.
A module to define a Pytorch dataset & dataloader from a set of annotations + images. For now, we can focus on the detection case.

Describe the solution you'd like
Taking as input an annotations dataframe and an image directory, add functionality to support and review a labelled dataset.

Describe alternatives you've considered
\

Additional context
\

[hrigar]

hi, i want to work on this , i will soon crate a draft PR, let me know if that works.

[hrigar]

I've implemented the basic dataset module that loads images and their annotations. For the detection part, we have two options:

1. Current Basic Implementation:

• Loads images from directory
• Associates with annotation dataframe

2. Should we extend to detection now?
   We could add:

• Bounding box handling
• Label processing
• Box format conversions

just want to make sure i am going in the right direction.
Let me know what do u think, open to changes and feedback.

[sfmig]

this xbatcher library looks great: https://github.com/xarray-contrib/xbatcher