AI Paper Atlas

Research map / Video / Video-language understanding

Scaling-up Perceptual Video Quality Assessment

Ziheng Jia, Zicheng Zhang, Xiaorong Zhu, Chunyi Li, Jinliang Han, Xiaohong Liu, Guangtao Zhai, Xiongkuo Min

AAAI 2026 · Video › Video-language understanding · approach: Video LLMs with fine-grained understanding · 8 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use complementary (sequential) training instead of direct training (ablation)OmniVQA models · Q-bench-video-test overallaccuracy (%): 58.33% -> 58.50%Improved
use complementary training instead of direct training (ablation)Rater · LSVQ (test)SRCC: 0.880 -> 0.902Improved
use complementary training instead of direct training (ablation)Chatter · OmniVQA-FG-Bench machine annotatedaccuracy (%): 68.00% -> 69.87%Improved
use complementary training instead of random mixing (ablation)Chatter · Q-bench-video-test overallaccuracy (%): 52.78% -> 58.50%Improved
use complementary training instead of random mixing (ablation)Rater · LSVQ (test)SRCC: 0.898 -> 0.902No clear effect
use complementary training instead of random mixing (ablation)Rater · KoNViD-1k, LIVE-VQC, OmniVQA-MOS-20K (test)SRCC: KoNViD-1k 0.887 -> 0.895; LIVE-VQC 0.822 -> 0.826; MOS-20K 0.840 -> 0.837Mixed
scale instruction data from 100k to 400k (ablation)Chatter · Q-bench-video-test, OmniVQA-FG-BenchaccuracyMixed
add human-annotated data to mixed training (ablation)Chatter · Q-bench-video-testaccuracyImproved
use chatter (understanding model) for quality rating instead of dedicated raterChatter · OmniVQA-MOS-20K (test)SRCC: 0.837 (Rater) vs 0.810Hurt
train on OmniVQA-Chat-400K with in-context dataChatter vs VQA2-Assistant · OmniVQA-FG-Bench machine annotatedaccuracy (%): 55.33% -> 69.87%Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy