Research map / Image generation / Autoregressive and latent generation
Scaling Up Personalized Image Aesthetic Assessment via Task Vector Customization
Jooyeol Yun, Jaegul Choo
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 |
|---|---|---|---|
| customize layer-wise task vectors from GIAA/IQA databases with trainable coefficients | PIAA with OpenCLIP ViT-L/14 · REAL-CUR cross-database 10-shot | SROCC: 0.495 -> 0.577 (vs MTCL) | Improved |
| use task vector method cross-database vs models trained on AADB | PIAA task vector customization · AADB 100-shot | SROCC: 0.622 -> 0.654 (MTCL trained on AADB vs ours) | Improved |
| task vector customization vs best prior method in-domain | PIAA · Flickr-AES 10-shot | SROCC: 0.667 -> 0.668 (MTCL vs ours) | No clear effect |
| keep task vectors frozen and train only coefficients instead of fine-tuning the best-fit model (ablation) | PIAA ViT-L/14 · REAL-CUR 10-shot | SROCC: 0.303 -> 0.577 | Improved |
| use layer-wise task vectors rather than one per task (ablation) | PIAA ViT-L/14 · REAL-CUR 10-shot | SROCC: 0.438 -> 0.572 | Improved |
| use Bradley-Terry rank loss instead of MSE loss (ablation) | PIAA ViT-L/14 · REAL-CUR 10-shot | SROCC: 0.553 -> 0.572 | Improved |
| adaptive softmax-SROCC coefficient initialization instead of uniform (ablation) | PIAA ViT-L/14 · REAL-CUR 10-shot | SROCC: 0.520 -> 0.577 | Improved |
| adaptive initialization instead of best-fit initialization (ablation) | PIAA ViT-L/14 · REAL-CUR 10-shot | SROCC: 0.464 -> 0.577 | Improved |
| increase the number of databases (task vectors) used | PIAA ViT-L/14 · REAL-CUR | SROCC | Improved |
| use smaller ViT-B/16 backbone instead of ViT-L/14 (ablation) | PIAA task vectors · REAL-CUR 10-shot | SROCC: 0.577 -> 0.562; FPS 94.41 -> 417.95 | Mixed |
Closest papers by meaning
- Advancing Comprehensive Aesthetic Insight with Multi-Scale Text-Guided Self-Supervised Learning (AAAI 2025 · similarity 0.89)
- Rethinking Personalized Aesthetics Assessment: Employing Physique Aesthetics Assessment as An Exemplification (CVPR 2025 · similarity 0.88)
- Bridging Cognitive Gap: Hierarchical Description Learning for Artistic Image Aesthetics Assessment (AAAI 2026 · similarity 0.88)
- Probabilistic Prompt Adaptation for Unified Image Aesthetics and Quality Assessment (CVPR 2026 · similarity 0.87)
- 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)
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.
