Figure 1
Exemplary illustration of type (top left) and style (bottom) using male statues of the Archaic and Classical periods as examples.
Figure 2
Comparative representation between (a) pixel-based segmentation and (b) GradCAM-generated visualization of regions significant for classification.
Table 1
Segmentation performance measured using Intersection over Union. Mean IoU values and standard deviations across the five evaluation folds are reported for each class.
| METRIC | HAIR | L. EAR | L. EYE | R. EYE | R. EAR | FACE | MOUTH | NOSE |
|---|---|---|---|---|---|---|---|---|
| IoU | 0.845 | 0.761 | 0.908 | 0.892 | 0.779 | 0.941 | 0.791 | 0.878 |
| Std | 0.040 | 0.102 | 0.013 | 0.016 | 0.094 | 0.011 | 0.068 | 0.044 |
Figure 3
Distribution of the IoU values across the five test folds for all classes combined (blue), hair (green) and left eye (red). The folds are separated by dashed lines.
Figure 4
Two segmentation examples of portraits of Augustus with IoU of 0.912 (a) and 0.885 (b), computed from a model trained on the same fold.
Figure 5
A portrait of Augustus in three-quarter view where the segmentation becomes worse, missing parts of the hair on the right side, failing to differentiate correctly between hair and face at the temples and labeling the garment over the left shoulder incorrectly as the left ear.
Figure 6
Segmentation results for the two emperors Caligula (a) and Tiberius (b). The test images differ from the training data, which contains only portraits of Augustus.
Figure 7
The normalized confusion matrix for the emperor classification task. Since the test predictions were aggregated across the five folds, each emperor class contains 24 images in total. Therefore, a single misclassification corresponds to approximately 0.04 in the normalized matrix.
Table 2
Emperor classification results obtained by stratified five-fold cross-validation. Values are reported as mean ± standard deviation across folds.
| METRIC | RESULT |
|---|---|
| Accuracy | 0.7821 ± 0.1046 |
| Macro Precision | 0.8249 ± 0.0737 |
| Macro Recall | 0.7850 ± 0.1066 |
| Macro F1 | 0.7807 ± 0.1065 |
Table 3
Per-class performance for emperor classification. Values are reported as mean ± standard deviation across folds.
| EMPEROR | PRECISION | RECALL | F1-SCORE |
|---|---|---|---|
| Augustus | 0.9333 ± 0.0913 | 0.8700 ± 0.1857 | 0.8851 ± 0.0914 |
| Tiberius | 0.6933 ± 0.2186 | 0.6700 ± 0.1718 | 0.6524 ± 0.1286 |
| Caligula | 0.7800 ± 0.1789 | 0.7600 ± 0.2191 | 0.7351 ± 0.1353 |
| Claudius | 0.8929 ± 0.1473 | 0.8400 ± 0.2191 | 0.8500 ± 0.1491 |
Figure 8
In (a) a portrait of Augustus is classified correctly as Augustus whereas in (b) another portrait of Augustus is erroneously classified as Tiberius.
Figure 9
Overview of GradCAM visualizations for representative emperor classification test images. The emperors are arranged in columns, starting with Augustus, followed by Tiberius, Caligula, and Claudius. Green borders indicate correct classifications, red borders indicate incorrect classifications.
Figure 10
The normalized confusion matrix for the type classification task. Since the test predictions were aggregated across the five folds, each class contains 17 images in total. Therefore, a single misclassification corresponds to approximately 0.06 in the normalized matrix.
Table 4
Portrait type classification results obtained by stratified five-fold cross-validation. Values are reported as mean ± standard deviation across folds.
| METRIC | RESULT |
|---|---|
| Accuracy | 0.7952 ± 0.2153 |
| Macro Precision | 0.8417 ± 0.1778 |
| Macro Recall | 0.8083 ± 0.1990 |
| Macro F1 | 0.7867 ± 0.2254 |
Table 5
Per-class performance for portrait-type classification. Values are reported as mean ± standard deviation across folds.
| TYPUS | PRECISION | RECALL | F1-SCORE |
|---|---|---|---|
| Louvre MA 1280 | 0.8500 ± 0.2236 | 0.7833 ± 0.3312 | 0.7657 ± 0.2691 |
| Prima Porta | 0.8333 ± 0.2357 | 0.8333 ± 0.2357 | 0.8076 ± 0.1935 |
Figure 11
The example images were generated using the presented GradCAM approach on different test fold samples and highlight the regions that are decisive for the classification. The types Prima Porta and Louvre MA 1280 are arranged in rows, with Prima Porta on top and Louvre MA 1280 below. Green borders indicate correct classifications (first three images in each row), red borders indicate incorrect classifications (last image in each row).