Blink estimation dataset + training code

[KevinCortacero]

This pull request provides a first version of the training code and blink dataset.

Some work is remaining:

• refactoring on the datasets
• a better way to load rt-bene dataset (pandas dataframe from csv?)

[KevinCortacero]

Thank you a lot for all your feedbacks @Tobias-Fischer :) I fixed most of the small stuff and many of the remaining things are necessary to speed up the tests and will be removed just after :) Regarding the class "RT_BENE", have you any suggestion?

[Tobias-Fischer]

Okay. When do you think you will finish the tests so we can remove the debugging stuff? I will go through the code once more after that is done.

I think RTBeneDataset would be a good class name.

[Tobias-Fischer]

More generally, I just noticed that the blink training code files are mixed with the gaze training code files, both in rt_gene_model_training. Please separate so that the blink stuff is in a separate folder, i.e. rt_bene_model_training.

[Tobias-Fischer]

Reminder for myself: Modify standalone readme file

[Tobias-Fischer]

Hi @ngageorange, I checked the code and it looks okay to me now. Would you mind having a look over it?

Remaining items for @Twarz:

• [ ] Retrain the models, particularly provide a MobileNet based version
• [ ] Provide code to evaluate the models

Remaining items for me:

• [x] Provide README files in the rt_bene_model_training and rt_bene_dataset folders

[Tobias-Fischer]

Great, thanks a lot for reviewing @ngageorange! Agreed with the merging of train_blink_model.py and blink_model_factory.py - fixed in https://github.com/Tobias-Fischer/rt_gene/pull/51/commits/d2be22e5b7f33a7131aa7345f86eb209e25e14f2

I also updated the README files.

I will merge this now and update #35 with a list of to-dos.