From Collection Drawers to AI - A One-Week Recipe
Arianna Salili-James · Bots, Bits, and Biodiversity
The short versionA useful AI-ready digitisation setup can be built in a week from recycled materials and borrowed cameras.
What this was about
Arianna Salili-James described a collaboration with NIBIO (Norway) in which a team built a digitisation rig in one week from recycled parts (a BugDome box, cheap lights, two borrowed cameras and a rotating platform, inspired by NHM's ALICE). They photographed bark beetle specimens and took 360° videos (useful for label reading), trained a YOLO model for genus- and species-level classification using GBIF and in-house images, and built 3D meshes from video with COLMAP for outreach. A refined three-camera version with one-button software and label extraction is in development; in Q&A she estimated 20-40 seconds per specimen and very low costs besides cameras.
Why it matters. Demonstrates a low-cost, rapid path to digitisation and AI classification for smaller collections.
In the room
- NIBIO's entomology collection has about 200,000 pinned insects; focus on Coleoptera (bark beetles).
- Plan: prototype rig, one week of photography, AI species identification, 3D reconstruction.
- Rig built from a BugDome box, foam, zip ties, cheap lights, two borrowed cameras and a rotating platform (inspired by ALICE).
- 360° videos gave excellent label views and may feed automatic label extraction; videos are temporary.
- GBIF data filtered by an AI tool plus NIBIO and NHM photos used to train a YOLO model for genus and species classification with good results.
- 3D meshes created from video using COLMAP for outreach.
- Refined three-camera box with one-button capture software and label extraction under development; the team seeks a name for it.
Notable moments
Transcript
Automatically generated captions can contain mistakes, especially in names and technical terms. Times are relative to the room recording.
Read the transcript ↓
Related talks.
Pinned insect digitization conveyor: image and data transcription workflows
An industrial conveyor workflow with QR-encoded metadata and automated skeletal records makes mass digitisation of pinned insects feasible.
Agentic AI Systems for Specimen Interaction and Camera-Based Observation
Off-the-shelf local vision language models can pick out and count specimen morphology and spot damage with minimal guidance.
Automated data transcription and Darwin Core field itemization from insect specimen labels using machine learning
Labellum goes beyond label OCR to itemise insect-label text into Darwin Core fields, with high accuracy for most fields but problems with catalogue numbers.