update command builder to match fine tuning script & add convenience scripts

[uoefeb]

This PR should be applied AFTER https://github.com/SanniM3/video_summarisation_git/pull/2

What this does:

• add some convenience scripts for setting everything up (downloads model, samples videos, etc)
• updates the "training_command_builder" to match the args in @SanniM3's modified fine tuning script
• adds some explanation to the README
• adds the annotations file (16mb... yuck...)

Please take a look and let me know if this is ok to merge.

---

As a side note, I'm still having issues getting get fine tuning script to run on GCP. It's complaining about not having a GPU at the moment, so it looks like something in my instance isn't configured right... If either of you have had this issue and resolved it, please let me know; it would be a big help.

[uoefeb]

@SanniM3, @ip342,

I've added a flag to allow us to toggle random caption selection on and off.

The random caption selector uses a fixed seed, so random captions should be reproducible.

The reason for having a choice for random selection is based on comments in this article (below), which was added to the official GIT readme after we forked from it.

Many image captioning datasets contain multiple captions per image. In those cases, a common strategy is to randomly sample a caption amongst the available ones during training.

If all of this looks fine to you, please feel free to hit the merge button on this PR. If not, leave some comments and I'll try to look at them during some downtime.