Personalized Image Aesthetic Assessment on LAPIS
0.573Spearman's RhoLinear-Hidden
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Linear-HiddenSupport=100-shot, Backbone=Qwen3-VL, Model Size=8B2026.04 | 0.573 | 0.313 | — | — | — | |
| Linear-HiddenSupport=100-shot, Backbone=Gemma 3, Model Size=12B2026.04 | 0.571 | 0.323 | — | — | — | |
| Linear-HiddenSupport=100-shot, Backbone=Qwen3-VL, Model Size=2B2026.04 | 0.568 | 0.321 | — | — | — | |
| Linear-HiddenSupport=100-shot, Backbone=Qwen3-VL, Model Size=4B2026.04 | 0.568 | 0.319 | — | — | — | |
| Linear-HiddenSupport=100-shot, Backbone=Gemma 3, Model Size=4B2026.04 | 0.568 | 0.328 | — | — | — | |
| Linear-Hidden (Reduce)Support=100-shot, Backbone=Qwen3-VL, Model Size=2B2026.04 | 0.48 | 0.224 | — | — | — | |
| Linear-Hidden (Reduce)Support=100-shot, Backbone=Gemma 3, Model Size=4B2026.04 | 0.469 | 0.22 | — | — | — | |
| Linear-Hidden (Reduce)Support=100-shot, Backbone=Qwen3-VL, Model Size=4B2026.04 | 0.468 | 0.202 | — | — | — | |
| Linear-Hidden (Reduce)Support=100-shot, Backbone=Qwen3-VL, Model Size=8B2026.04 | 0.459 | 0.197 | — | — | — | |
| Linear-Hidden (Reduce)Support=100-shot, Backbone=Gemma 3, Model Size=12B2026.04 | 0.446 | 0.189 | — | — | — | |
| Linear-Hidden (GIAA)Support=100-shot, Backbone=Qwen3-VL, Model Size=4B2026.04 | 0.42 | 0.148 | — | — | — | |
| Linear-Hidden (GIAA)Support=100-shot, Backbone=Qwen3-VL, Model Size=8B2026.04 | 0.42 | 0.151 | — | — | — | |
| Linear-Hidden (GIAA)Support=100-shot, Backbone=Qwen3-VL, Model Size=2B2026.04 | 0.418 | 0.148 | — | — | — | |
| Linear-Hidden (GIAA)Support=100-shot, Backbone=Gemma 3, Model Size=12B2026.04 | 0.416 | 0.155 | — | — | — | |
| Linear-Hidden (GIAA)Support=100-shot, Backbone=Gemma 3, Model Size=4B2026.04 | 0.413 | 0.153 | — | — | — | |
| P-MLLMSize=8B, Zero-shot=true, Profile=Demographic2026.04 | 0.413 | — | 0.029 | 0.411 | 0.028 | |
| P-MLLM-SSize=8B, Zero-shot=true, Profile=Demographic2026.04 | 0.339 | — | 0.022 | 0.339 | 0.027 | |
| GPT-4o-miniZero-shot=true, Profile=Demographic2026.04 | 0.302 | — | 0.032 | 0.293 | 0.025 | |
| Few-shotSupport=10-shot, Backbone=Qwen3-VL, Model Size=8B2026.04 | 0.264 | 0.48 | — | — | — | |
| AesExpertSize=7B, Zero-shot=true, Profile=Demographic2026.04 | 0.251 | — | 0.032 | 0.253 | 0.031 | |
| Raw TextBackbone=Gemma 3, Model Size=12B2026.04 | 0.233 | 1.335 | — | — | — | |
| Adjust-BiasSupport=100-shot, Backbone=Gemma 3, Model Size=12B2026.04 | 0.233 | 0.442 | — | — | — | |
| Few-shotSupport=10-shot, Backbone=Gemma 3, Model Size=12B2026.04 | 0.227 | 0.459 | — | — | — | |
| Few-shotSupport=10-shot, Backbone=Qwen3-VL, Model Size=4B2026.04 | 0.221 | 0.38 | — | — | — | |
| LoRASupport=100-shot, Backbone=Gemma 3, Model Size=12B2026.04 | 0.201 | 1.022 | — | — | — | |
| Raw TextBackbone=Qwen3-VL, Model Size=4B2026.04 | 0.176 | 0.937 | — | — | — | |
| Adjust-BiasSupport=100-shot, Backbone=Qwen3-VL, Model Size=4B2026.04 | 0.176 | 0.231 | — | — | — | |
| Raw TextBackbone=Qwen3-VL, Model Size=8B2026.04 | 0.175 | 0.763 | — | — | — | |
| Adjust-BiasSupport=100-shot, Backbone=Qwen3-VL, Model Size=8B2026.04 | 0.175 | 0.206 | — | — | — | |
| LoRASupport=100-shot, Backbone=Qwen3-VL, Model Size=8B2026.04 | 0.164 | 1.386 | — | — | — | |
| LoRASupport=100-shot, Backbone=Qwen3-VL, Model Size=4B2026.04 | 0.153 | 1.58 | — | — | — | |
| Qwen2.5-VLSize=7B, Zero-shot=true, Profile=Demographic2026.04 | 0.145 | — | 0.03 | 0.154 | 0.028 | |
| Few-shotSupport=10-shot, Backbone=Qwen3-VL, Model Size=2B2026.04 | 0.142 | 1.265 | — | — | — | |
| Few-shotSupport=10-shot, Backbone=Gemma 3, Model Size=4B2026.04 | 0.127 | 0.354 | — | — | — | |
| Raw TextBackbone=Gemma 3, Model Size=4B2026.04 | 0.119 | 1.34 | — | — | — | |
| Adjust-BiasSupport=100-shot, Backbone=Gemma 3, Model Size=4B2026.04 | 0.119 | 0.162 | — | — | — | |
| LoRASupport=100-shot, Backbone=Gemma 3, Model Size=4B2026.04 | 0.116 | 0.936 | — | — | — | |
| Q-InstructSize=7B, Zero-shot=true, Profile=Demographic2026.04 | 0.101 | — | 0.024 | 0.095 | 0.017 | |
| Raw TextBackbone=Qwen3-VL, Model Size=2B2026.04 | 0.098 | 0.778 | — | — | — | |
| Adjust-BiasSupport=100-shot, Backbone=Qwen3-VL, Model Size=2B2026.04 | 0.098 | 0.264 | — | — | — | |
| LoRASupport=100-shot, Backbone=Qwen3-VL, Model Size=2B2026.04 | 0.026 | 0.701 | — | — | — |