Add Transform for DetCon + MultiViewTransformV2 for torchvision.transforms.v2

[liopeer]

Changes

• added MultiViewTransformV2 as a new base class for all future transforms requiring torchvision.transforms.v2
• added DetConSimCLRViewTransform as the single view transform for DetCon
• added DetConSTransform that uses the previous DetConSimCLRViewTransform to generate 2 views
  • the applied transforms differ only in their application of Gaussian blurring
  • to compare the exact hyperparams from the model, check the paper (DetCon) on page 9/13
  • add MultiViewTransformV2 and DetConSTransform to docs

    How has it been tested?

• testing correct tensor shapes
• testing that masks are unaffected if user-provided
• testing that masks are changed (and correct) if not user-provided

  Docs Screenshots (they do not show all the constructor arguments, but they are there!)

  

[codecov[bot]]

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Project coverage is 85.12%. Comparing base (f33a90f) to head (25ed891). Report is 1 commits behind head on master.

Additional details and impacted files

```diff @@ Coverage Diff @@ ## master #1737 +/- ## ========================================== + Coverage 85.03% 85.12% +0.08% ========================================== Files 159 161 +2 Lines 6631 6667 +36 ========================================== + Hits 5639 5675 +36 Misses 992 992 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

---

🚨 Try these New Features:

• Flaky Tests Detection - Detect and resolve failed and flaky tests