ArtSleuth
Computational Art Analysis Framework
WikiArt Benchmark
81,444 images · 27 styles · 129 artists · 11 genres
| Backbone | Style | F1 | Artist | Top-5 | Genre |
|---|---|---|---|---|---|
| DINOv2 · ViT-B/14 | 57.5% | 0.553 | 64.7% | 90.9% | 71.0% |
| CLIP · ViT-L/14 | 67.1% | 0.656 | 74.6% | 95.9% | 75.0% |
| Fusion · frozen | 65.0% | 0.633 | 71.0% | 94.2% | 74.2% |
| Fusion · fine-tuned | 71.6% | 0.703 | 77.8% | 96.2% | 75.1% |
| Fusion · end-to-end | 72.7% | — | 79.0% | 96.9% | 76.6% |
Top four rows: linear probes. Bottom row: end-to-end heads. Fine-tuning: SupCon + CE, 3-block unfreeze, cosine annealing, 5 epochs, Tesla P100. All numbers macro-averaged.