DOVER (Disentangled Objective Video Quality Evaluator, ICCV 2023) is a modern deep learned no-reference video-quality model built for user-generated content, the phone clips and webcam videos that arrive with no pristine original to compare against. Its distinguishing idea is to split a UGC clip's quality into two separate scores, a technical score for distortions such as blur, noise, and compression, and an aesthetic score for composition and appeal, rather than collapsing both into one number. It looks only at the decoded pixels and is among the accuracy leaders for blind video quality, reaching roughly 0.83 to 0.88 SROCC with human scores on in-distribution large UGC datasets, and it ships a lightweight DOVER-Mobile variant. Its catch is the one all learned blind models share: accuracy collapses cross-dataset, so a model validated on social clips can rank surveillance or screen content backwards. It needs a GPU, carries a research licence, and every score should be reported as a band, not a point.
The handbook · 7 volumes
Fora Soft Video Engineering Handbook
DOVER scores user-generated video without a reference and separates technical quality from aesthetics, which helps when uploads have no pristine source. We wrote seven volumes covering the whole video stack, and for DOVER start with Volume 1: no-reference and perceptual metrics and how compression damage shows up, with Volume 6 on models, inference and deployment. Written by Nikolay Sapunov, CEO at Fora Soft.
Need to score user uploads without a reference — would DOVER work for you?
Send us a sample of your UGC: we run DOVER and other no-reference models, compare them to viewer ratings and recommend one with its compute cost. 250+ video products shipped since 2005.
