Quote:
Originally Posted by werecow
Interesting! I have a background in AI and for a while I've been thinking about doing a data science project involving sword collecting. A friend from the forums has also indicated that he's interested in joining in.
I'd be particularly interested in seeing if we can create a system for automated recognition of either images of swords, or even better, images of marks and decorations, or at least an image search for similar images that could be of help in identifying the markings in question. But right now I'm not sure where to get good training data (it would require a pretty extensive annotated database of images and metadata).
Anyway, it's not well thought out at the moment but when it takes a more concrete form I might contact you.
|
Thank you for your interest! Recent advances provide an exciting opportunity to combine traditional cultural heritage documentation with AI/ML capabilities. The semantic web work done by museums and cultural institutions over the past decades has created a strong foundation of structured data and standardized vocabularies that we can build upon.
While AI has become incredibly powerful at extracting structure from unstructured data, having well-organized source data is still crucial for training effective models. The CIDOC CRM ontology and Getty vocabularies I'm using provide exactly the kind of rich, structured relationships that can help train and validate AI systems.
I've spent time fine-tuning both vision and text models using LORA adapters, and I'd be very interested in collaborating on applying these techniques to weapon classification and mark identification. The standardized documentation approach I'm developing could help provide the kind of consistent, well-labeled training data that's often the biggest challenge in these projects.
Let me know if you'd like to discuss this further as your ideas take shape. There could be some interesting opportunities to combine our work.