Automatic tool to help instrument/bow recognition

[Juju93]

Hi everyone,
A friend passionated about old violins and bows and I are creating a photo-based tool to help violin and bows identification.

We're sure you'll enjoy the concept and would love to get your insights about it.

Could you take this 1 minute anonymous survey ?

Your feedback will help us hit all the right notes.

Thank you !

Survey

 

 

 

 

 

 

 

Edited July 24, 2025 by Juju93

[IBK]

Need more info on  your concept, but don't think that it is possible.

[Juju93]

Hi IBK
We are currently developing an AI to recognize instruments and bows, as well as tree rings (to do automatic dendro reports). What we are trying to know with our survey is : do you have interest in such a product ?
The link to our survey was broken, I just edited my post to add it properly.
Please find it again here : https://violin-intelligence.typeform.com/to/JrlgwLu1

[GeorgeH]

Identification of historical makers by AI will never be possible due to insurmountable problems with the data set.

[Violadamore]

1 hour ago, Juju93 said:

Hi IBK
We are currently developing an AI to recognize instruments and bows, as well as tree rings (to do automatic dendro reports). What we are trying to know with our survey is : do you have interest in such a product ?
The link to our survey was broken, I just edited my post to add it properly.
Please find it again here : https://violin-intelligence.typeform.com/to/JrlgwLu1

I meddled with a similar idea (not AI, granted, but a dichotomous key), about 10 years ago, and quietly gave up on it.  The concept runs aground to start with on the number of (real, not label names, and mostly anonymous) historical violin makers, plus how many of them (GDG is an example) seem to have changed styles radically over time (or from day to day.....).  This is compounded by the millions of trade violins made up from parts (an endless march of Frankenfiddles), the thousands of makers who ran multi-employee shops, legions of talented fakers, the corruption of the historical record (trusting online photo archives?  Don't!), etc., etc. .     

Not to discourage you, but I feel that your project is not only doomed, but may generate some bad information of its own before you pull the plug on it.  Don't just get the code running, and rush to publish.  

[martin swan]

2 hours ago, Juju93 said:

Hi IBK
We are currently developing an AI to recognize instruments and bows, as well as tree rings (to do automatic dendro reports). What we are trying to know with our survey is : do you have interest in such a product ?
The link to our survey was broken, I just edited my post to add it properly.
Please find it again here : https://violin-intelligence.typeform.com/to/JrlgwLu1

Whose dataset are you planning to steal in order to create your own database?

Any fully reliable photo archive will not be public domain.

[Pluto Vermin]

The two major patterns are Strad-ish and Guarnieri-ish.  Have it just assign fiddles to those two categories first, as evidence you aren't hallucinating yourself.  Or just do that with f-holes

[Juju93]

Hello everyone,
Thank you for your insights, all your comments are precious to me.
At this stage, I am disseminating our market survey to get to know if there is interest in such a project.
In our opinion, the AI current developments are offering the opportunity to make it happen.
Thanks again for your time and attention

[GeorgeH]

On 7/15/2025 at 9:03 AM, Juju93 said:

A friend passionated about old violins and bows and I are creating a photo-based tool to help violin and bows identification.

In your statement above, what does "identification" encompass?

3 hours ago, Juju93 said:

In our opinion, the AI current developments are offering the opportunity to make it happen.

I'm also curious as to what is your experience in either AI or violin identification to support this opinion?

[Blank face]

One should consider that human experts more often than not don't agree on identifications (except the apparently most easy ones), and/or that they might change with time. This flaws are also reflected in databases. So the first decision would be made between the reliability of sources, or it will inevitably end like this: https://www.theguardian.com/technology/2025/jul/09/grok-ai-praised-hitler-antisemitism-x-ntwnfb

[Juju93]

Good morning,

By « identification » I meant finding leads on a possible maker, school, model, date - for a violin or a bow.I’m not saying AI will replace an expert eye. But I think I can come up with new tool that can help anyone having an instrument / bow in hands, to save time in their authentication process. And I hope I will soon be able to come back here with a trial version so you can make up your mind.

As I said, I am only here to try to get insights about if such an app would be of any interest to you.

My question here is : If such a solution was within hand reach, would you use it ?Thank you

Edited July 18, 2025 by Juju93

[martin swan]

5 hours ago, Juju93 said:

 

My question here is : If such a solution was within hand reach, would you use it ?Thank you

It's a good question.

Who would use such an app?

Anyone who has accrued significant professional knowledge of violins and bows would do a better job than AI ... I can't see myself ever using such an app and in relative terms I'm not even that knowledgeable.

The knowledge itself is commercially very significant, so acquiring this knowledge (or the appearance of it) without too much hard work/genuine study would seem most relevant to ''cash in the attic" speculators looking at things at auction or maybe violin shops and household clearance firms who see things that are beyond their pay grade.

How are you going to charge for this knowledge? Are you really going to hand over the algorithm that tells someone that the violin for sale for £50 at their local flea market is worth £500,000? 

And how will you indemnify the users of your app who overpay for things based on inaccurate readings from your app?

A dendro app would be a different concept, but I can't see how it would be commercially viable, given that you'd have to buy someone else's extremely valuable database ... it's not like getting a dendro done is very expensive.

[GeorgeH]

9 hours ago, Juju93 said:

finding leads on a possible maker, school, model, date - for a violin or a bow.

That is what I thought. Violin identification relies on examination inside and outside the instruments from many different angles, subtle measurements, evaluating repairs, varnishes, labels, etc. There is no data set available nor will there ever be that could be used to teach an AI algorithm to identify makers, schools, models, dates, etc.

[Blank face]

It’s only necessary to study Maestronet UD threads to recognize how difficult an identification by photos can be, how many pictures, views from different angles and details, informations about inside structures, speculations about alterations and much more is necessary to come to any significant conclusions. 

[Will Turner]

As AI can only operate via photographs, there's an intrinsic level of detail that is missed.    Forgeries can easily be mistaken via photograph, even with photocopied labels that the AI models likely would have trouble distinguishing from the originals.

Until it's a robotic luthier tied to the AI that handles the instrument and bow, I don't believe this is possible.    Again, even professional humans will provide inaccurate attestations that can only be based on partial or missing details and provenance. 

There's even still active debates over some violins being Stradivari's or an incredible copy made by Vuillaume.  

[Juju93]

Thank you everyone for your insights. 

I know that violin identification is a very delicate, subjective an complex task. 
You are absolutely right when you say it requires much more than 3 pictures to authenticate a violin. There are tons of makers, some well documented, some not. 
It will be easier for the ones who made hundreds of instruments, and impossible for the obscur ones. 
Nevertheless, it may be possible to narrow down the leads in most cases, by giving matching percentages. 
Correct pictures shows numerous information : outlines, shapes, type of wood, purflings, growth rings etc... 

Bow identification will be easier thanks to reliable data available for each maker. 

Since AI can compare a picture from a CCTV recording and a social media profile pic, it will be possible to identify a single instrument on 2 different pictures. Of course it's possible to make a bench copy of an instrument or a bow and impossible to clone a human being. That's true, and that's why no AI will replace human expertise.

I hope this message will help you better understand my researches.

[GeorgeH]

3 hours ago, Juju93 said:

Bow identification will be easier thanks to reliable data available for each maker. 

No, it won’t. It has even less probability of success than violin identification.

3 hours ago, Juju93 said:

You are absolutely right when you say it requires much more than 3 pictures to authenticate a violin.

3 hours ago, Juju93 said:

Since AI can compare a picture from a CCTV recording and a social media profile pic, it will be possible to identify a single instrument on 2 different pictures.

Do you see the contradiction between these two statements?

[Blank face]

12 hours ago, Juju93 said:

Correct pictures shows numerous information : outlines, shapes, type of wood, purflings, growth rings etc..

Sadly it seems you don’t have much clue about violin identification and nothing about bows.

[bungling_amateur]

On 7/23/2025 at 8:45 AM, Juju93 said:

Since AI can compare a picture from a CCTV recording and a social media profile pic, it will be possible to identify a single instrument on 2 different pictures. Of course it's possible to make a bench copy of an instrument or a bow and impossible to clone a human being. That's true, and that's why no AI will replace human expertise.

 

12 hours ago, Blank face said:

Sadly it seems you don’t have much clue about violin identification and nothing about bows.

It sounds to me like Juju93 is using the word "identification" in a different way from the rest of us. It almost sounds like facial recognition - we don't expect the system to view a previously unseen human, and then predict his or her name and address. Facial recognition systems look at a newly-presented human face, and compare it to a large database of named reference faces, and find the best match.

So perhaps this system will have a large reference collection of violin photographs, and then if I go to the Ashmolean and take a snap of the Messiah with my phone camera, and upload my snapshot to the system, it will successfully match my snap against its reference images to tell me that my snap shows the Messiah?

Such a system could be useful for the police to identify a recovered stolen violin (if it had previously been registered on the system and reference images provided)

[martin swan]

46 minutes ago, bungling_amateur said:

It sounds to me like Juju93 is using the word "identification" in a different way from the rest of us. It almost sounds like facial recognition - we don't expect the system to view a previously unseen human, and then predict his or her name and address. Facial recognition systems look at a newly-presented human face, and compare it to a large database of named reference faces, and find the best match.

So perhaps this system will have a large reference collection of violin photographs, and then if I go to the Ashmolean and take a snap of the Messiah with my phone camera, and upload my snapshot to the system, it will successfully match my snap against its reference images to tell me that my snap shows the Messiah?

Such a system could be useful for the police to identify a recovered stolen violin (if it had previously been registered on the system and reference images provided)

This is very like the skill of some individuals who recognize certain violins from myriad clues, grain appearance, model, varnish colour, blemishes and damages etc.

I can conceive of an AI app that would recognize a specific violin but which would be incapable of putting a name to an unknown violin. This is of course not expertise but photo memory.

With bows the margin of error would be much greater since it's pretty easy to make a true copy.

[Moray]

I look forward to the day when all these AI recognition tools create their own forum (MAIstronet?) and start having terse discussions about makers, blockology and "The Usuals". 

[GeorgeH]

On 7/23/2025 at 3:45 AM, Juju93 said:

Of course it's possible to make a bench copy of an instrument or a bow and impossible to clone a human being.

The technology to clone a human already exists. It is called somatic cell nuclear transfer (SCNT), and it has been used to clone several different types of mammals.

[Blank face]

4 hours ago, bungling_amateur said:

It sounds to me like Juju93 is using the word "identification" in a different way from the rest of us. It almost sounds like facial recognition

Taken her word it doesn’t seem so.

On 7/18/2025 at 10:05 AM, Juju93 said:

By « identification » I meant finding leads on a possible maker, school, model, date - for a violin or a bow

 

[lFred]

At least now we know with whom you'll be sharing all your violins archives https://violin-intelligence.com/

[Dr. Mark]

On 7/24/2025 at 5:34 AM, martin swan said:

This is of course not expertise but photo memory

Arguably, that's what AI is however it's marketed.