Medical MVP: VQA viability

[andreimano]

We should check if it's possible to solve some VQA tasks on medical data. VQA is especially nice when combined with #43 and #35 .

Depends on #33 .

[andreimano]

MONAI seems to have no pretrained models for this task yet 1.

This paper 2 reports decent results when using deep convolutional network coupled with transformers or recurrent neural networks.

A bigger dataset can be found at 3, but it's a lot newer.

The example in 4 is from "Show, attend and tell", which is a pretty old paper. A simple CNN encoder - RNN decoder could be used as a baseline, it should also work fine with Captum. 3 reports good results with this type of network architecture.

[andreimano]

A demo should be possible, we should first train a baseline and make sure that it works, and then mix it up with Captum. See #45.