Classes for echocardiography datasets

[mxochicale]

🚀 Feature

Currently, the annotated datasets only are for 4CV which require to create some sort of binary class (4CV and BGR) where 4CV is four chamber view and BRG is background.

Motivation

Pitch

Annotated 4CV videos are rendered into image frames to be cropped and masked and we would like to have:

• [x] a binary labels in the data-loader for 4CV and BRG labels
• [x] Linked to vital-ultrasound/ai-assisted-echocardiography-for-low-resource-countries#14, a sampling method will be refined here considering NUMBER_OF_SEGMENTS, FRAGMENTS_PER_SEGMENT, and RANDOM_FRAMES_PER_SEGMENT as show in the following figure.
• [x] Potentially add a third class for semi-4CV (or something like entering and leaving 4CV)

Alternatives

Additional context

Few references that might be of help to this ticket:

• https://pytorch.org/vision/master/auto_examples/plot_video_api.html#sphx-glr-auto-examples-plot-video-api-py
• https://github.com/vital-ultrasound/lung/blob/main/multiclass_pytorch/datasets/LUSVideoDataset.py
• https://pytorch.org/tutorials/beginner/data_loading_tutorial.html sample = {'image': image, 'landmarks': landmarks}
• https://medium.com/analytics-vidhya/creating-a-custom-dataset-and-dataloader-in-pytorch-76f210a1df5d
• Video-Dataset-Loading-Pytorch: https://github.com/RaivoKoot/Video-Dataset-Loading-Pytorch Added: Mon 13 Dec 09:36:41 GMT 2021
• https://torchvideo.readthedocs.io/en/latest/samplers.html added Fri 28 Jan 08:52:50 GMT 2022

[mxochicale]

Few points to consider for future work:

• Potentially, we would like to add a third class:
  • soft-class
  • semi-4CV (or something like entering and leaving 4CV)