Add Vision Transformer as alternative to Res Net

[Tarek-Sallam]

Summary

Currently the representation network is modelled as a ResNet. It would be interesting to see if we could get better model performance or faster more efficient training by using a Vision Transformer, as in recent years the Vision Transformer has proved to be a more popular alternative to ResNet.

Implementation Notes

Share any technical ideas, code, or pseudocode:

# Example or pseudocode

Additional Context

Add links, screenshots, or other relevant info.

Acceptance Criteria

• Clear, testable criteria for completion