Research map / Video / Video segmentation and tracking
Learning Flexible Generalization in Video Quality Assessment by Bringing Device and Viewing Condition Distributions
Nikolay Safonov, Dmitriy S. Vatolin
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add viewing-condition adaptation module on top of LPIPS | LPIPS · held-out phone model 1 | Kendall rank correlation: 0.308 -> 0.652 | Improved |
| add viewing-condition adaptation module on top of VMAF | VMAF · held-out phone model 1 | Kendall rank correlation: 0.509 -> 0.600 | Improved |
| add adaptation module on top of AVQT | AVQT · held-out phone model 1 | Kendall rank correlation: 0.439 -> 0.694 | Improved |
| add adaptation module on top of PSNR | PSNR · held-out phone model 3 | Kendall rank correlation: 0.501 -> 0.430 | Hurt |
| add adaptation module on top of NLPD | NLPD · held-out phone models 2 and 3 | Kendall rank correlation: 0.468 -> 0.415 | Hurt |
| add adaptation module on top of MUSIQ | MUSIQ · held-out phone models 1-3 | Kendall rank correlation: 0.320 -> 0.320 | No clear effect |
| add adaptation module to all metrics, evaluated on held-out devices | 32 IQA/VQA metrics · three most frequent phone models | Kendall rank correlation | Mixed |
| use condition pool: sample viewing conditions independently of annotations to augment training (ablation) | adaptation module · held-out devices | Kendall rank correlation | Improved |
| use Blade-Chest aggregation conditioned on viewing conditions instead of Bradley-Terry | score aggregation · dataset scores | SROCC: 0.881 (vs BT-scores) | No clear effect |
| use tanh activations in networks instead of linear (ablation) | Blade-Chest networks · dataset | score fit | Improved |
| evaluate on HDR-capable vs SDR displays for HDR video | subjective scores · HDR videos | PLCC vs reference: 0.881 -> 0.933 | Improved |
| larger screen size group of devices | crowd subjective scores · agreement with desktop scores | PLCC: 0.909 -> 0.945 | Improved |
Closest papers by meaning
- Analysis of Video Quality Datasets via Design of Minimalistic Video Quality Models (TPAMI 2024 · similarity 0.87)
- Exploring Video Quality Assessment on User Generated Contents from Aesthetic and Technical Perspectives (ICCV 2023 · similarity 0.87)
- Modular Blind Video Quality Assessment (CVPR 2024 · similarity 0.85)
- KVQ: Boosting Video Quality Assessment via Saliency-guided Local Perception (CVPR 2025 · similarity 0.84)
- Learned Scanpaths Aid Blind Panoramic Video Quality Assessment (CVPR 2024 · similarity 0.83)
- FineVQ: Fine-Grained User Generated Content Video Quality Assessment (CVPR 2025 · similarity 0.83)
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.
