Table 12 Extended metrics of the proposed ArtFusionNet model, and the comparison models.
From: Enhancing artistic style classification through a novel ArtFusionNet framework
Model | AUC-ROC | Accuracy (%) | Per-Class Recall | Cohen’s Kappa | F1-Score | Validation Loss |
|---|---|---|---|---|---|---|
Transformer-only Model | 0.82 | 79.5 | Style 1: 0.77 Style 2: 0.74 Style 3: 0.71 | 0.75 | 0.7823 | 1.1254 |
CLIP-based Framework | 0.96 | 95.4 | Style 1: 0.90 Style 2: 0.88 Style 3: 0.86 | 0.93 | 0.9850 | 0.1875 |
Swin Transformer | 0.95 | 92.3 | Style 1: 0.87 Style 2: 0.85 Style 3: 0.83 | 0.92 | 0.9801 | 0.2150 |
CNN-only Model | 0.85 | 81.7 | Style 1: 0.80 Style 2: 0.78 Style 3: 0.75 | 0.80 | 0.8139 | 1.0426 |
DeiT | 0.97 | 96.3 | Style 1: 0.91 Style 2: 0.87 Style 3: 0.85 | 0.94 | 0.9875 | 0.1601 |
Proposed model | 0.98 | 99.00 | Style 1: 0.92 Style 2: 0.89 Style 3: 0.87 | 0.95 | 0.9906 | 0.1227 |
