Bugfix - same output for PIL and tensor when centercrop size is greater than imgsize

[saurabheights]

This PR fixes #3297

[vfdev-5]

Looks good, thanks @saurabheights !
A nit comment to improve the code a bit

[codecov]

Codecov Report

Merging #3333 (9ccb744) into master (859a535) will decrease coverage by 0.01%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #3333      +/-   ##
==========================================
- Coverage   73.90%   73.88%   -0.02%     
==========================================
  Files         104      104              
  Lines        9618     9624       +6     
  Branches     1544     1546       +2     
==========================================
+ Hits         7108     7111       +3     
- Misses       2028     2030       +2     
- Partials      482      483       +1     

Impacted Files	Coverage Δ
torchvision/transforms/transforms.py	83.68% <ø> (-0.43%)	⬇️
torchvision/transforms/functional.py	81.73% <100.00%> (+0.25%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 859a535...9ccb744. Read the comment docs.

[vfdev-5]

Looks good, thanks @saurabheights !
Just a nit remark about randomness in the tests. I know that it was already there and you replicated that, but from a practical point of view it will be hard to debug random parameters... Anyway, this improvement can be addressed in future.

[saurabheights]

@vfdev-5 Just setting even_image_size to (20, 20) should be enough. It still tests all possibilities(even odd img/crop sizes, different size diff). If you want, I can push that change.

Also, regarding drop in coverage, not 100% sure but I think it comes from other tests testing center crop method but not using test condition of cropsize>imgsize. However, adding test conditions to other tests would take me some time.

[vfdev-5]

@saurabheights let's ignore codecov failure.

Just setting even_image_size to (20, 20) should be enough. It still tests all possibilities(even odd img/crop sizes, different size diff). If you want, I can push that change.

Yes, I understand. Anyway, let's do it in another PR. My remark about both random params: even_image_size and delta.