The MOS provided is subjectively collected or pseudo ? about dvqa HOT 4 CLOSED

Yes. The provided subjective score is SUR (Satisfied User Ratio) , which is derived from JND-VideoSet. Traditional MOS score is in a scale of [1, 5] and 1 for worst and 5 for best. However, 'SUR' is defined in a different way.

VideoSet contains JND samples in terms of QP values. SUR is defined as SUR = 1 - CDF(JND) along the QP axis. Clearly, it gives the percentage of users that are unable to tell the difference between reference and distorted videos.

SUR is mentioned in Section 7 of VIdeoSet and explicitly defined in Section 2 of SUR prediction.

from dvqa.

@tommyhq Thanks. So, you presumed that videos at the 1st JND share a same quality score.

from dvqa.

@Sissuire I am not clear about your concern.

Note that SUR has nothing to do with the 75% JND location. It uses all raw JND samples.

On the other hand, we got 1 JND sample from 1 volunteer during subjective test. That JND location is the boundary between perceptual lossless and lossy (with high probability), for that specific volunteer. Thus, it is assumed that the quality of the 1st JND (raw samples) is well aligned.

from dvqa.

@tommyhq Thank you for the helpful reply, which havs resolved my doubt.

from dvqa.