Bug: non squared input images break SklearnClassification task

[AlessandroPolidori]

Summary

Describe the purpose of the pull request, including:

• What problem does it solve?
  Non squared input images currently brake sklearn classification task in two ways:
  1- torchscript export fails
  2- gradcam fails
• How does it solve the problem?
  1-?
  2- Now mask is reconstructed correctly even when width and height are different.
• Remaining issues
  1- I have not tried to solve this problem yet.
  2- The mask adjustment is not a trivial task. The reshape() function needs integers, so you need to find two values, height and width, such that w*h = total_size of the mask. But there is no way to know a priori how to truncate height and width values. The solution that satisfied me the most (based on output gradcams) implies a reduction of mask values. I did different tests and the resulting gradcams seem good.

Type of Change

Please select the one relevant option below:

• Bug fix (non-breaking change that solves an issue)

Checklist

Please confirm that the following tasks have been completed:

• I have tested my changes locally and they work as expected. (Please describe the tests you performed.)
• I have added unit tests for my changes, or updated existing tests if necessary.
• I have updated the documentation, if applicable.
• I have installed pre-commit and run locally for my code changes.

Screenshots or Visuals (Optional)

If applicable, please provide screenshots, diagrams, graphs, or videos of the changes, features or the error.

[lorenzomammana]

pytorch/pytorch#82068

Related to torchscript fail, we can't do anything about it right now as it is dino code