Automate progressive image resizing

[yellowcap]

To train the multi size images architecture, we need to either stop/restart the model every ~10 epochs, changing progressively the image size in the data loader.

Preferrably we would implement a pytorch hook to make this change automatically.

[yellowcap]

Good idea but not on the roadmap for now.

[srmsoumya]

Note: While doing multinode with slurm, we have to adjust num_tasks, num_workers while changing the chip_size, this was not trivial to implement, will skip this for now.