fixed Gamma casting

[sanketsans]

Fixes #3067

Removed the double casting of the result to the original dtype.
Now, the casting is only done if the original dtype is different from float32 type which is used to adjust the gamma of the image / tensor.

Since the method requires the image dtype to be in float format. If not, then line no. 400 converts it to the float format. Then needs just one check to change it back to its original format. If it is already in float format then we don't need to change it.

vision/torchvision/transforms/functional_tensor.py

Lines 397 to 406 in 9e71fda

	result = img
	dtype = img.dtype
	if not torch.is_floating_point(img):
	result = convert_image_dtype(result, torch.float32)
	result = (gain * result ** gamma).clamp(0, 1)
	result = convert_image_dtype(result, dtype)
	result = result.to(dtype)
	return result

[datumbox]

@sanketsans Please avoid opening new PRs in relation to this, so that we can keep track of the discussions in one place (for future reference previously related PRs are #3470, #3471).

As you can see your current PR breaks the unit-tests. This is because the convert_image_dtype() does more than converting the image to specific dtype and thus can't be replaced with a simple to(). As discussed at #3067 (comment) and #3471, the only necessary modification for fixing the issue is just to remove the unnecessary to() call.

If you decide to update this PR with the requested change, please ping me to review it again. Thanks!

[sanketsans]

@datumbox Sorry about multiple PRs. I will make sure to avoid it in the future.
I have committed the necessary changes as you mentioned. You can review it.
Thanks

[codecov]

Codecov Report

Merging #3472 (6f5b065) into master (f637c63) will decrease coverage by 0.00%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #3472      +/-   ##
==========================================
- Coverage   78.70%   78.69%   -0.01%     
==========================================
  Files         105      105              
  Lines        9705     9704       -1     
  Branches     1555     1555              
==========================================
- Hits         7638     7637       -1     
  Misses       1576     1576              
  Partials      491      491              

Impacted Files	Coverage Δ
torchvision/transforms/functional_tensor.py	79.19% <ø> (-0.04%)	⬇️

---

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 f637c63...6f5b065. Read the comment docs.