This seems to be an issue in the new version of the algorithm. Certain pairs of images can produce negative ssimulacra2 scores, which I'm assuming shouldn't be possible. For example, the following source/distorted image pair produces a score of -109.
This seems to be an issue in the new version of the algorithm. Certain pairs of images can produce negative ssimulacra2 scores, which I'm assuming shouldn't be possible. For example, the following source/distorted image pair produces a score of -109.