Research map / Vision-language and multimodal / Multimodal LLMs
Beyond Absolute Scores: Relative Edit-induced Difference for Generalizable Image Aesthetic Assessment
Qifei Jia, Xintong Yao, Minghao Li, Yajie Chai, Qiming Lu, Baoyue Shen, Yasen Zhang, Runyu Shi, Ying Huang, Yue Zhang
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 |
|---|---|---|---|
| train Aes-R1 on RED-20k instead of AesCoT-15k (ablation) | Aes-R1 · AVA/FLICKR/TAD66K average | SRCC: 0.5947 -> 0.6465 | Improved |
| use consistent group (source and its edits) instead of shuffled batch (ablation) | RED-Aes RL · AVA/FLICKR/TAD66K average | SRCC: 0.6532 -> 0.6758 | Improved |
| add content edits alongside quality edits (ablation) | RED-Aes · AVA/FLICKR/TAD66K average | SRCC: 0.6421 -> 0.6758 | Improved |
| add quality edits alongside content edits (ablation) | RED-Aes · AVA/FLICKR/TAD66K average | SRCC: 0.6315 -> 0.6758 | Improved |
| add content edits to RL stage (ablation) | RED-Aes · AVA/FLICKR/TAD66K average | SRCC: 0.6543 -> 0.6758 | Improved |
| add contrastive objective PT-O1 to SFT (ablation) | SFT · AVA/FLICKR/TAD66K average | SRCC: 0.5577 -> 0.5809 | Improved |
| add edit causal learning PT-O2 (ablation) | SFT with PT-O1 · AVA/FLICKR/TAD66K average | SRCC: 0.5809 -> 0.6218 | Improved |
| use relative-ranking-consistency RL instead of Aes-R1 RL (ablation) | cold start base · AVA/FLICKR/TAD66K average | SRCC: 0.6465 -> 0.6758 | Improved |
| add RL on cold start base (ablation) | cold start base · AVA/FLICKR/TAD66K average | SRCC: 0.5907 -> 0.6758 | Improved |
| increase editing models from 1 to 5 (one judge) (ablation) | data construction · AVA/FLICKR/TAD66K average | SRCC: 0.5939 -> 0.6335 | Improved |
| increase editing models from 7 and judges 1 to 3 (5+3) (ablation) | data construction · AVA/FLICKR/TAD66K average | SRCC: 0.6335 -> 0.6758 | Improved |
| increase judges from 3 to 5 (ablation) | data construction · AVA/FLICKR/TAD66K average | SRCC: 0.6758 -> 0.6762 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- UniPercept-8B — UniPercept: Towards Unified Perceptual-Level Image Understanding across Aesthetics, Quality, Structure, and Texture (ICML 2026) · better on 1 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
- Unlocking the Essence of Beauty: Advanced Aesthetic Reasoning with Relative-Absolute Policy Optimization (ICLR 2026 · similarity 0.92)
- Bridging Cognitive Gap: Hierarchical Description Learning for Artistic Image Aesthetics Assessment (AAAI 2026 · similarity 0.90)
- Advancing Comprehensive Aesthetic Insight with Multi-Scale Text-Guided Self-Supervised Learning (AAAI 2025 · similarity 0.88)
- ArtiMuse: Fine-Grained Image Aesthetics Assessment with Joint Scoring and Expert-Level Understanding (CVPR 2026 · similarity 0.87)
- ELTA: An Enhancer against Long-Tail for Aesthetics-oriented Models (ICML 2024 · similarity 0.87)
- Aligning Vision Models with Human Aesthetics in Retrieval: Benchmarks and Algorithms (NeurIPS 2024 · similarity 0.85)
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.
