LIDAR reveals an 11-kilometre lost Mayan causeway under the Petén canopy

Three months of airborne LIDAR over the Petén basin in Guatemala has resolved 11.4 km of paved sacbé connecting two minor Classic-period centres to Tikal. The data passes that produced the result, the vegetation-filter parameters that almost hid it, and what comes next.

2026-05-22 lidartikalmayafieldworkguatemalaphotogrammetry

Training a transformer on Linear B: what 1,427 tablets do and don't teach a 90-million-parameter model

A small transformer trained from scratch on the Mycenaean Linear B corpus produces three credible readings for an undeciphered fragment of the KN Fp series. The model is wrong in interesting ways. Here's the architecture, the training data, and what the failures tell us about epigraphy.

2026-05-11 transformerlinear-bmycenaeanepigraphydecodingknossos

Walking through the Karnak hypostyle hall in VR: what the model gets right, and what it can't

A 134-pillar VR reconstruction of the Karnak hypostyle hall, built from 47,000 photogrammetric source images and the Ramessid-era polychromy survey, is now running at 90 Hz on a Quest 3. The mesh is faithful to within 4 mm. The lighting is the open question — and it is a harder question than the geometry.

2026-05-04 vrphotogrammetrykarnakegyptreconstructionpolychromy

Troubleshooting: Present audio and video with IIIF

Troubleshooting common problems when you present audio and video with IIIF: diagnose the usual errors, find the root cause fast and apply fixes that actually

2026-04-30 aviiifmultimediatroubleshootingerrorsfixes

The ethics of digitally reconstructing destroyed heritage

When the Umayyad Mosque minaret in Aleppo was reduced to rubble in 2013, photogrammetric models were our last record of what stood there. We can rebuild it in a renderer to sub-centimetre accuracy. Whether we should — and on whose authority — is a question the technology will not answer for us.

2026-04-30 ethicsreconstructionaleppopalmyraheritagephotogrammetry

Troubleshooting: Model signed and hostile ties

Troubleshooting common problems when you model signed and hostile ties: diagnose the usual errors, find the root cause fast and apply fixes that actually hold

2026-04-27 signedalliancesnetworkstroubleshootingerrorsfixes

Reconcile place names in your data: A Practical Guide

A practical guide to reconcile place names in your data for working historians and archivists, covering the workflow end to end with concrete examples you can

2026-04-27 reconciliationplacescleaningguidepracticalworkflow

Best Practices to Sustain a crowdsourcing community

Best practices and a working checklist to sustain a crowdsourcing community, so your results stay consistent, documented and defensible across a whole

2026-04-27 sustainabilitycommunitycrowdsourcingbest-practiceschecklistquality

Best Practices to Test a format migration safely

Best practices and a working checklist to test a format migration safely, so your results stay consistent, documented and defensible across a whole collection

2026-04-22 migrationtestingqualitybest-practiceschecklisthistory

A photogrammetry workflow for fragile manuscripts that won't touch the page

Most photogrammetric documentation pipelines assume an object that can be moved. Bound or charred manuscripts cannot. The workflow we have settled on at the Bodleian — sub-millimetre mesh density without a single page-turn — relies on a turntable that never moves the page, a calibrated 60 megapixel camera, and a quiet patience with Agisoft Metashape.

2026-04-18 manuscriptphotogrammetrymetashapebodleianconservationpreservation

1234 ... 58