[RavSinghChandan]

tensor_from_numpy turns a grayscale image of shape (H, W) into a (1, H, W) tensor with the default dtype=torch.float32 (through to_tensor), but with torch.float16 or torch.uint8 it permuted the array as if it had a channel axis and raised:

RuntimeError: permute(sparse_coo): number of dimensions in the tensor input does not match the length of the desired ordering of dimensions

tensor_from_pil with a grayscale image and one of these dtypes failed the same way. A grayscale array now gets a channel axis first, so it gives a (1, H, W) tensor with every dtype, like it already did with float32. Color images are unchanged

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.74%. Comparing base (2e274dc) to head (773543d).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2201      +/-   ##
==========================================
- Coverage   97.75%   97.74%   -0.01%     
==========================================
  Files         170      170              
  Lines       10974    10975       +1     
==========================================
  Hits        10728    10728              
- Misses        246      247       +1     

Flag	Coverage Δ
unittests	97.74% <100.00%> (-0.01%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[felixdittrich92]

Thanks 👍