Research map / Vision-language and multimodal / Human and embodied modeling
Fine-grained Image Aesthetic Assessment: Learning Discriminative Scores from Relative Ranks
Zhichao Yang, Jianjie Wang, Zhixianhe Zhang, Pangu Xie, Xiangfei Sheng, Pengfei Chen, Leida Li
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 fine-grained ranking training to coarse-only training (ablation) | FGAesQ · FGAesthetics | Series (s-Acc, s-SRCC)/2: 0.364 -> 0.600 | Improved |
| add coarse-grained pre-training to fine-only training (ablation) | FGAesQ · FGAesthetics | Series (s-Acc, s-SRCC)/2: 0.299 -> 0.600 | Improved |
| use two-stage pretrain then joint alternating training instead of sequential Coarse | Fine (ablation) | FGAesQ · FGAesthetics | Series (s-Acc, s-SRCC)/2: 0.380 -> 0.600 | Improved |
| add DiffToken (ablation) | FGAesQ · FGAesthetics | Series (s-Acc, s-SRCC)/2: 0.423 -> 0.600 | Improved |
| add DiffToken (ablation) | FGAesQ · FGAesthetics | Pair (Acc, F1)/2: 0.666 -> 0.753 | Improved |
| add CTAlign comparative text alignment (ablation) | FGAesQ · FGAesthetics | Series (s-Acc, s-SRCC)/2: 0.581 -> 0.600 | Improved |
| add RankReg instead of direct ranking training (ablation) | FGAesQ · FGAesthetics | Series (s-Acc, s-SRCC)/2: 0.571 -> 0.600 | Improved |
| add DiffToken at inference (w DiffToken vs w/o) | FGAesQ · FGAesthetics Natural | series s-SRCC: 0.664 -> 0.729 | Improved |
| fine-tune existing IAA model with ranking loss on FGAesthetics | Charm · AVA | SRCC: 0.777 -> 0.470 | Hurt |
| fine-tune existing IAA model with ranking loss on FGAesthetics | Charm · FGAesthetics | Series (s-Acc, s-SRCC)/2: 0.419 -> 0.477 | Improved |
| fine-tune NIMA with ranking loss | NIMA · AVA | SRCC: 0.633 -> 0.330 | Hurt |
| use larger MLLM-based scorer instead of traditional deep model | Q-Align vs Charm · FGAesthetics Natural | series s-SRCC: 0.404 -> 0.496 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- Q-Align — Q-Align: Teaching LMMs for Visual Scoring via Discrete Text-Defined Levels (ICML 2024) · better on 4 of 4 shared results
- Charm — CHARM: Collaborative Harmonization Across Arbitrary Modalities for Modality-Agnostic Semantic Segmentation (AAAI 2026) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- ArtiMuse: Fine-Grained Image Aesthetics Assessment with Joint Scoring and Expert-Level Understanding (CVPR 2026 · similarity 0.91)
- Advancing Comprehensive Aesthetic Insight with Multi-Scale Text-Guided Self-Supervised Learning (AAAI 2025 · similarity 0.89)
- FiVA: Fine-grained Visual Attribute Dataset for Text-to-Image Diffusion Models (NeurIPS 2024 · similarity 0.89)
- ELTA: An Enhancer against Long-Tail for Aesthetics-oriented Models (ICML 2024 · similarity 0.88)
- Alchemist: Turning Public Text-to-Image Data into Generative Gold (NeurIPS 2025 · similarity 0.87)
- Bridging Cognitive Gap: Hierarchical Description Learning for Artistic Image Aesthetics Assessment (AAAI 2026 · similarity 0.87)
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.
