Research map / Image generation / Autoregressive and latent generation
Life-IQA: Boosting Blind Image Quality Assessment through GCN-enhanced Layer Interaction and MoE-based Feature Decoupling
Long Tang, Huiyu Duan, Guoquan Zheng, Jianbo Zhang, Jie Hao, Liang Yuan
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 MoE head at end of decoder (vs FFN) (ablation) | Life-IQA · KADID-10K | SROCC: 0.924 -> 0.940 | Improved |
| add MoE head (vs removing it) (ablation) | Life-IQA · CSIQ | SROCC: 0.934 -> 0.966 | Improved |
| add query GCN instead of multi-head attention (ablation) | Life-IQA · KADID-10K | SROCC: 0.917 -> 0.940 | Improved |
| add query GCN (vs removing it) (ablation) | Life-IQA · CSIQ | SROCC: 0.952 -> 0.966 | Improved |
| add stage3 key/value interaction path (ablation) | Life-IQA · CSIQ | SROCC: 0.938 -> 0.966 | Improved |
| use Life-IQA decoder instead of vanilla Transformer decoder (ablation) | Swin Transformer · KADID-10K | SROCC: 0.926 -> 0.940 | Improved |
| scale Swin backbone from Tiny to Base (ablation) | Life-IQA · LIVEC | SROCC: 0.875 -> 0.896 | Improved |
| use only shallow stages (stage1/2) for quality regression | pretrained backbones · LIVEC | SROCC | Hurt |
| Life-IQA vs VISGA | Life-IQA · LIVE | SROCC: 0.987 -> 0.982 | Hurt |
| Life-IQA vs FsPN in cross-dataset LIVE | Life-IQA · LIVE (train KonIQ) | SROCC: 0.829 -> 0.822 | Hurt |
| Life-IQA vs DEIQT with 20% training data | Life-IQA · LIVEC | SROCC: 0.792 -> 0.804 | Improved |
Closest papers by meaning
- Distilling Spatially-Heterogeneous Distortion Perception for Blind Image Quality Assessment (CVPR 2025 · similarity 0.86)
- Integrating Global Context Contrast and Local Sensitivity for Blind Image Quality Assessment (ICML 2024 · similarity 0.85)
- Towards Syn-to-Real IQA: A Novel Perspective on Reshaping Synthetic Data Distributions (NeurIPS 2025 · similarity 0.85)
- Adaptive Feature Selection for No-Reference Image Quality Assessment by Mitigating Semantic Noise Sensitivity (ICML 2024 · similarity 0.85)
- Feature Denoising Diffusion Model for Blind Image Quality Assessment (AAAI 2025 · similarity 0.84)
- Bridging the Synthetic-to-Authentic Gap: Distortion-Guided Unsupervised Domain Adaptation for Blind Image Quality Assessment (CVPR 2024 · similarity 0.84)
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.
