Visual Question Answering on HGI SemiosisArt
93.2AccuracyHuman (Expert)
Evaluation results
| Method | Links | |
|---|---|---|
| Human (Expert)Group=Human Reference, Backbone=-2026.04 | 93.2 | |
| SemJudge (Gemini-Flash)Group=Semiosis Quality, Backbone=Zero-shot MLLM2026.04 | 92.4 | |
| SemJudge (Qwen-35B-A3B)Group=Semiosis Quality, Backbone=Zero-shot MLLM2026.04 | 91 | |
| SemJudge (Qwen-9B)Group=Semiosis Quality, Backbone=Zero-shot MLLM2026.04 | 86.1 | |
| Human (Non-expert)Group=Human Reference, Backbone=-2026.04 | 81.5 | |
| ArtCoTGroup=Formal Analysis, Backbone=Zero-shot MLLM2026.04 | 80.4 | |
| ArtiMuseGroup=Art Interpretation, Backbone=Tuned MLLM2026.04 | 67.7 | |
| LMM4LMMGroup=Rationale&Scoring, Backbone=Tuned MLLM2026.04 | 44 | |
| GalleryGPTGroup=Art Interpretation, Backbone=Tuned MLLM2026.04 | 26.1 | |
| Random GuessGroup=Conventional Scorers, Backbone=-2026.04 | 25.2 |