[FFmpeg-devel] [PATCH 10/11] tests/swscale: constrain reference SSIM for low bit depth formats

[Niklas Haas]

From: Niklas Haas <git at haasn.dev>

Sometimes, the reference SSIM is significantly higher than the
SSIM level expected for the test. This is the case when the source format
has a much lower bit depth than the destination format. In this case, the fact
that legacy swscale does not accurately preserve the source dither pattern
gives it an unfair advantage in a direct comparison, leading to false
positives.

For example, conversion like rgb4 -> rgb565 should be lossless, but swscale
low passes / downscales the input chroma, throwing away massive amounts of
detail. This gives it a higher SSIM score since the lowpassed result removes
some of the dither noise that was present in the source.
---
 libswscale/tests/swscale.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/libswscale/tests/swscale.c b/libswscale/tests/swscale.c
index bce495db90..117ed2144e 100644
--- a/libswscale/tests/swscale.c
+++ b/libswscale/tests/swscale.c
@@ -321,6 +321,18 @@ static int run_test(enum AVPixelFormat src_fmt, enum AVPixelFormat dst_fmt,
             goto error;
 
         get_ssim(ssim_sws, out, ref, comps);
+
+        /* Legacy swscale does not perform bit accurate upconversions of low
+         * bit depth RGB. This artificially improves the SSIM score because the
+         * resulting error deletes some of the input dither noise. This gives
+         * it an unfair advantage when compared against a bit exact reference.
+         * Work around this by ensuring that the reference SSIM score is not
+         * higher than it theoretically "should" be. */
+        if (src_var > dst_var) {
+            const float src_loss = (2 * ref_var + c1) / (2 * ref_var + src_var + c1);
+            ssim_sws[0] = FFMIN(ssim_sws[0], src_loss);
+        }
+
         ssim_ref = ssim_sws;
     }
 
-- 
2.48.1