single precision support in skimage.metrics module

[grlee77]

Description

#5219 should be reviewed first since this PR is built on top of that one (to reuse the same _float_dtype utility).

Computations of means as in mse were intentionally set to use dtype=float64 and that is preserved here to keep good accuracy on larger images. Similarly, for structural_similarity, I added single precision support in most of the function, but kept the final mean computation to generate mssim in double precision.

Checklist

• Docstrings for all functions
• Gallery example in ./doc/examples (new features only)
• Benchmark in ./benchmarks, if your changes aren't covered by an
  existing benchmark
• Unit tests
• Clean style in the spirit of PEP8

For reviewers

• Check that the PR title is short, concise, and will make sense 1 year
  later.
• Check that new functions are imported in corresponding __init__.py.
• Check that new features, API changes, and deprecations are mentioned in
  doc/release/release_dev.rst.

[pep8speaks]

Hello @grlee77! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2021-07-15 08:58:04 UTC

[rfezzani]

I just left some minor comments, otherwise, everything is fine 😉 thank you @grlee77.

[rfezzani]

👍

[mkcor]

Thanks, @grlee77: I'm only this far (tiny gesture) from approving/merging!

[mkcor]

Weird... at b94bce2 test_PSNR_float[float16] is failing because of line 69 in skimage/metrics/tests/test_simple_metrics.py; I fetched your branch and saw the same thing locally; to have tests pass, I had to do:

diff --git a/skimage/metrics/tests/test_simple_metrics.py b/skimage/metrics/tests/test_simple_metrics.py
index a69b9abb3..edcb28fd4 100644
--- a/skimage/metrics/tests/test_simple_metrics.py
+++ b/skimage/metrics/tests/test_simple_metrics.py
@@ -66,7 +66,7 @@ def test_PSNR_float(dtype):
     with expected_warnings(['Inputs have mismatched dtype']):
         p_mixed = peak_signal_noise_ratio(cam / 255.,
                                           np.float32(cam_noisy / 255.))
-    assert_almost_equal(p_mixed, p_float64, decimal=5)
+    assert_almost_equal(p_mixed, p_float64, decimal=decimal)

But here on GitHub line 69 reads assert_almost_equal(p_mixed, p_float64, decimal=decimal) 🤔
Rerunning all CI workflows just in case... Have you done any crazy rebasing? 😅

[mkcor]

Although GitHub would display otherwise, the code base definitely had assert_almost_equal(p_mixed, p_float64, decimal=5), apparently:

That's why I took the liberty of committing 508415e, which passed the CI tests.