| <p>Artifex estimates what is <strong>typical of its reference collection</strong>. The limitations below bound what its outputs can mean.</p> | |
| <h2>Scientific</h2> | |
| <ul> | |
| <li><strong>Correlation, not attribution.</strong> A linear classifier on image embeddings learns statistical regularities of the | |
| training images. It has no notion of materials, technique, iconography or provenance, and it can be right for the | |
| wrong reasons (frames, mounts, paper tone, photographic set-up, image resolution).</li> | |
| <li><strong>Confounds from photography and cataloguing.</strong> Each museum photographs and colour-manages its collection differently. | |
| Collection-disjoint experiments estimate how much performance depends on such institution-specific cues; a large drop | |
| means the in-collection numbers overstate what to expect for arbitrary photographs.</li> | |
| <li><strong>Calibration is distribution-specific.</strong> Temperature scaling and abstention thresholds are fitted on validation works | |
| from the same two collections. For phone photos, reproductions in books, or traditions absent from the collection, | |
| the scores are not calibrated.</li> | |
| <li><strong>Closed label set.</strong> The model can only output supported categories. A work from an unsupported tradition will be | |
| assigned to the nearest supported one (the interface warns about this, and the out-of-distribution flag catches only | |
| some such cases).</li> | |
| <li><strong>Abstention protects overall accuracy, not small classes.</strong> The threshold is chosen to reach 80 % selective accuracy on | |
| validation works overall. For broad regions this target is met without abstaining (Europe and East Asia dominate), so | |
| the region head never abstains on confidence, even though recall for small classes is much lower than for the large | |
| ones (see the per-class table on the Experiments page).</li> | |
| <li><strong>Ablations are diagnostic.</strong> Greyscale, cropping and low-level-baseline comparisons show sensitivities; they do not | |
| explain <em>what</em> the model uses.</li> | |
| <li><strong>Small classes and wide intervals.</strong> Overall test metrics of the main experiment carry 95% confidence intervals from a | |
| grouped bootstrap; they are wide because the test split holds a few hundred leakage groups. Per-class metrics for | |
| categories with few test works have even wider uncertainty and no intervals are computed for them.</li> | |
| <li><strong>One split, one estimate.</strong> The website's Experiments page shows a single grouped split. Re-drawing the split | |
| matters: successive builds in October 2026 (removing 21 non-paintings; grouping manuscripts by painter rather than | |
| text author) moved single-split region macro-F1 between 0.689 and 0.776, and on the final build alone ten re-drawn | |
| splits give 0.716 ± 0.024 (range 0.686–0.751), with Himalayan F1 between 0.39 and 0.86. Repeated-split results are | |
| produced by <code>scripts/evaluate_multisplit.py</code> and reported in the paper (<code>paper/</code>).</li> | |
| </ul> | |
| <h2>Data and historical</h2> | |
| <ul> | |
| <li><strong>Collection bias.</strong> Two American encyclopaedic museums, filtered to CC0 images, do not represent world painting. Holdings | |
| over-represent European oil painting, Chinese and Japanese scroll painting and South Asian court painting; they | |
| under-represent Africa, Oceania, the Americas before 1500, Southeast Asia and most vernacular or religious traditions. | |
| Twentieth-century works are scarce because most are still in copyright.</li> | |
| <li><strong>Court and élite art.</strong> Museum holdings favour élite and collectable objects; the "visual signature" of a region in | |
| Artifex is largely the signature of what Western museums collected.</li> | |
| <li><strong>Label provenance.</strong> Region labels come from museum culture fields or, failing that, an artist's recorded nationality. | |
| The latter is a school attribution, not a place of production. Museum attributions themselves change with scholarship.</li> | |
| <li><strong>Taxonomy choices.</strong> Broad regions are coarse and partly arbitrary at their edges (Himalayan vs. Indian Buddhist painting; | |
| Mughal vs. Persianate; colonial Latin American painting). Ambiguous records are excluded from training, which may bias | |
| classes towards clear-cut examples.</li> | |
| <li><strong>Date uncertainty.</strong> Many dates are broad ranges ("18th century") or approximate. Works spanning several bins have no | |
| period label; bins are coarse; exact-bin accuracy penalises boundary cases.</li> | |
| <li><strong>Manuscripts and albums.</strong> Folios are photographed and catalogued individually; grouping by accession lot and capping | |
| each group reduces, but may not eliminate, related-work leakage (e.g. dispersed albums held under different lots or by | |
| both museums).</li> | |
| <li><strong>Copies and reproductions.</strong> Later copies dated to their originals, or works "in the style of" a tradition, are labelled | |
| as the museum records them.</li> | |
| </ul> | |
| <h2>Technical</h2> | |
| <ul> | |
| <li><strong>Frozen encoder.</strong> DINOv2 was trained on natural photographs; it was not adapted to paintings. Fine-tuning is future work.</li> | |
| <li><strong>Input format.</strong> Every image is letterboxed to 224 × 224. Very wide handscrolls or tall hanging scrolls are reduced to | |
| thin strips, losing detail; some handscroll images were too small to keep at all.</li> | |
| <li><strong>Out-of-distribution detection</strong> is a nearest-neighbour distance heuristic, calibrated so ~1 % of in-distribution | |
| validation images are flagged. It is not a reliable detector of non-paintings.</li> | |
| <li><strong>Single photograph.</strong> Analysis uses one image; details, reverse sides and technical imaging are not considered.</li> | |
| <li><strong>In-browser edition.</strong> The website runs an ONNX export of the encoder with float16-stored weights and float16 | |
| reference embeddings. On held-out test images its predictions matched the Python application's exactly, and the | |
| similar-work rankings differed for at most 1 % of images; but image decoding is done by the visitor's browser: unusual | |
| files (CMYK JPEGs, 16-bit PNGs) may decode slightly differently from Pillow. Neither edition applies embedded colour | |
| profiles.</li> | |
| <li><strong>Snapshot data.</strong> The museums' APIs change; a rebuild later may differ. The manifest, dataset version and split files | |
| identify each run exactly.</li> | |
| </ul> | |
| <h2>Ethical</h2> | |
| <ul> | |
| <li>Artifex must not be used to authenticate, value or attribute objects for sale.</li> | |
| <li>Predictions about cultural origin can carry political and legal weight (e.g. provenance and restitution questions). | |
| They are not evidence in such contexts.</li> | |
| </ul> |