Error about varying image sizes in fully supervised model training

[hummuscience]

I am on the main branch and using the dev pip install.

Running the training hydra script as usual and get the following error:

From what I understood, this error should not occur when using a fully supervised model (without pca or temporal losses).

This is related to the training of the Topvieemouse5k dataset, which has varying sizes: #158

Btw, this error does not occur on the dynamic_crop branch.

[hummuscience]

I am a bit confused, as I was able to run this on the main brain a few days ago without getting this error. The only thing that changed was that I pulled the recent changes to the repo. So the error must have been quite recently, but I might be mistaken.

I switched to the dynamic_crop branch since it looks like this dependency on image size was merged there: #117

[themattinthehatt]

Oh yes, my apologies - ok I recommend switching to the unsuper-multiview branch as this (1) has the ability to train on multiple image sizes; and (2) does not run the validation data through the augmentation pipeline.

This branch is now PR #187 which will be merged into main in the next day or so!