A pottery drawing needs more than a cleaner outline

PyPottery’s drawing editor gives an inferred extension of a broken edge its own layer. That small feature explains a larger aim of archaeological software: producing an illustration whose separate decisions remain available for someone else to inspect.[5]

Developed by Lorenzo Cardarelli, the suite connects scanned field sheets, AI-assisted image cleaning, editable drawings and publication layouts. Following a drawing through those stages shows what must travel with it.[1]

Keep the label with the object

At the scanning stage, the researcher marks a vessel drawing and the text that belongs to it. A label can sit elsewhere on the sheet: it is explicitly attached to the selected drawing. The software reads those marked regions, then presents the label image beside an editable transcription. Inventory numbers, contexts and notes can therefore be checked against their source before export.[3]

The project retains the scans, crops, annotations and text corrections. A clean illustration can move into a catalogue without becoming detached from the writing that identified the object. This is a useful way to judge the workflow: ask whether the original relationship survives each transformation.[3]

Separate the parts before tracing

PyPotteryTrace helps select a region with points or a box, then assigns it a category such as profile, handle or decoration. Its handled-vessel example shows why this matters: a selection can catch the body and handle together. The operator excludes the handle from the profile, then saves it as a separate component.[4]

The resulting curves and their control points can remain editable in a vector drawing program, with components on named layers. Detail can also remain a raster image, represented by pixels. Vectorisation does not require turning every mark into the same kind of line.[4]

A reconstructed continuation is another distinct component. The editor estimates an arc from points near a break, falling back to a straight extension when those points align. It stores the result in a Reconstruction layer that can be omitted from export. The illustrator can inspect or remove that proposed continuation without redrawing the surviving edge.[5]

Where the 103 minutes went

Cardarelli’s October preprint reports one complete run on 240 pottery drawings from 50 sheets associated with Terramara di Montale, in Italy. The author used a desktop with an RTX 3070 Ti graphics card. Adding the durations in the paper’s Table 3 gives this breakdown:[2]

That is about 140 drawings an hour for this run. Manual and mixed work occupied roughly 60% of the elapsed time. The paper’s separate 40-fold median speedup comes from five experts’ estimates of time saved, not a timed comparison with a traditional workflow. Vectorisation required manual intervention for 21 of the 240 drawings.[2]

Publish the relationships as well as the shapes

At the layout stage, spreadsheet fields can determine order and grouping. Drawings from one category can start on a new page, with captions carrying their accompanying data. A grid keeps a regular reading order; the space-saving Puzzle layout can change positions within a page to make the shapes fit.[6]

Scale remains a measurement decision. The software calculates its scale bar from the supplied pixels-per-centimetre value and the chosen image multiplier. An incorrect scan resolution produces an incorrect bar even if the page looks tidy. Its SVG output keeps captions, numbers and the scale bar editable.[6]

The finished plate is therefore more than a collection of cleaned silhouettes. A researcher can organise it around an excavation context or a vessel type, retain the labels that support that grouping and adjust the presentation later. The drawing becomes useful because its relationships remain legible.[6]

Report an error or suggest a correction