The detail-loss metric (ADM, also written DLM) is a full-reference quality feature that separates compression damage into two distinct effects rather than lumping them together: detail that was lost, which makes the picture less clear, and impairment that was added, the extra junk that distracts the eye. It scores the detail loss on its own, which is what lets it cleanly catch blur and the smearing-away of texture — distortions that reduce clarity without necessarily adding visible noise. ADM is one of the elementary features fused inside VMAF, where it sits alongside Visual Information Fidelity (VIF) and a motion feature, and a trained regression model learns how heavily to weigh each on different content. The separation of lost detail from added impairment is its conceptual contribution: by not confusing the two, it gives the fused metric a cleaner signal about one kind of damage. Like the rest of VMAF's features it is full-reference, computed per frame, and meaningful only inside the trained fusion, not as a standalone verdict.
