Prieure de Sion Posted July 18 · Member Share Posted July 18 Perhaps some of you collectors have already tried the AI-Provenance search on Coincabinet.io (3 days free trial). The aim is to find provenances using AI algorithms. Collectors have already tried the search in the German forum, but only with poor results. Someone has just posted a good search result in a Facebook group - a reason for me to give it a try. Information can be found here: https://coincabinet.io/en/provenance_ai To search, you have to create an account and register. You can test the AI search for 3 days free of charge. The first thing I did was look for my new Caracalla denarius. If it hadn't been found, it would have been okay. What I find bad is that the results that were shown don't even have a slight resemblance! I would have preferred the result to be “we haven't found anything”! But look at the hits on the reverse - how can a coin with a horse be a possible match? And no - the young Caracalla does not look like Vespasian. 1 2 2 1 Quote Link to comment Share on other sites More sharing options...
Prieure de Sion Posted July 18 · Member Author Share Posted July 18 Ok, the website says that, among other things, Aurei in particular are scanned and indexed. Then I'll give my new Septimius Severus Aureus a try. In my thread, a user manually found some auctions where my Septimius Severus Aureus was sold / auctioned in the past without any problems. So it should be no problem for the AI to find something here. But the search did not find a single correct hit! It's really very easy to do (manually) with this Aureus. Sol compared to a standing Jupiter? Septimius Severus is like a Campgate? 1 1 Quote Link to comment Share on other sites More sharing options...
Restitutor Posted July 18 · Administrator Share Posted July 18 I'm shocked they are already attempting to monetize this when the returns are so bad. A mass public beta test is the smartest thing for them to do and only start charging once the returns are significantly more accurate. A 3 day free trial seems incredibly short-sighted but I guess time will tell. Also interested in where they are getting their data from. Did they purchase these images outright? I tried to get some auction results data from a few key houses for my own personal use, and got ghosted by all of them! Curious how he's able to sell this information. 2 5 Quote Link to comment Share on other sites More sharing options...
Prieure de Sion Posted July 18 · Member Author Share Posted July 18 Has anyone else tried this AI search and actually had success? Somehow the database still seems to me to be quite meagre and the search / hit evaluation is also completely faulty. As written - I would understand if coin A and coin B differed in a few things in the hits. But how can a standing person with a person on a horse be considered a “possible hit” - or the portrait of Septimius Severus and that of a Campgate? 1 Quote Link to comment Share on other sites More sharing options...
Prieure de Sion Posted July 18 · Member Author Share Posted July 18 1 minute ago, Restitutor said: I'm shocked they are already attempting to monetize this when the returns are so bad. A mass public beta test is the smartest thing for them to do and only start charging once the returns are significantly more accurate. A 3 day free trial seems incredibly short-sighted but I guess time will tell. I'm also shocked that the owners want around $70/year for these results. I would rather extend the trial period and continue to improve and collect more data - and then later make the business model a monthly subscription. 1 2 Quote Link to comment Share on other sites More sharing options...
Prieure de Sion Posted July 18 · Member Author Share Posted July 18 Yeah! And if you want delete your personal data and your account - you will get this, after you press the button - I will delete! Oh I "like" such Websites like this... Quote Link to comment Share on other sites More sharing options...
Prieure de Sion Posted July 18 · Member Author Share Posted July 18 Ok, come on - it's kind of fun to get these chaotic hits. Let's be fair and give the tool a third chance. My Gordianus Aureus. Ahhhhh Gordianus has a beard, no Jupiter is sitting, no Jupiter is Virtus. No - this is still completely immature - it's not even remotely helpful. But now I don't feel like it anymore... 2 Quote Link to comment Share on other sites More sharing options...
Salomons Cat Posted July 18 · Member Share Posted July 18 (edited) I agree that the provenance search seems ineffective. The model is not fully trained yet. If a particular coin is in their database, the chance of finding it appears to be only slightly above 75% and results appear random if a coin is not in their database. They emphasize pre-1970 catalogs, which is a good idea. They state that they don't charge for provenance searches unless a match is found. The creator provides more information here, which answers most questions: https://coincabinet.io/de/provenance_ai After reviewing the site, it does seem like a legitimate project, not a scam. As the name "Coincabinet" suggests, the service aims to provide a virtual coin cabinet. It is more of a competitor to colleconline.com rather than acsearch, coinarchives, or Coryssa. Therefore, what you are paying for is the virtual coin cabinet, not the (currently immature) AI search. Edited July 18 by Salomons Cat 1 1 Quote Link to comment Share on other sites More sharing options...
Deinomenid Posted July 18 · Supporter Share Posted July 18 I tried various coins a few weeks ago from different periods with known provenance and got comical results. Well the first few were comical then it just got irritating. Even if this is a sideshow to providing a coin cabinet experience it was a waste of my time. Time sinks professing to be more are not my thing. I did though end up using it to take quick and easy trimmed (whatever the term is) photos of the coins without the backgrounds. Quote Link to comment Share on other sites More sharing options...
Deinomenid Posted July 18 · Supporter Share Posted July 18 Ahh, I see the site's owner just posted about how embarrassed he is! That's something I suppose.... Quote Link to comment Share on other sites More sharing options...
Sulla80 Posted July 18 · Supporter Share Posted July 18 I thought it woudl be fun to try - it is just not ready for any real use: 1 1 1 Quote Link to comment Share on other sites More sharing options...
Harry G Posted July 18 · Member Share Posted July 18 I like the concept, but I just gave it a try and it wasn't exactly successful... With any luck, it'll get better as the AI gets more training. For example, Google reverse image search used to be terrible when trying to identify coins, but now it works almost all the time for me 1 1 1 Quote Link to comment Share on other sites More sharing options...
CPK Posted July 18 · Supporter Share Posted July 18 What a joke! Quote Link to comment Share on other sites More sharing options...
Sulla80 Posted July 18 · Supporter Share Posted July 18 It gets worse - I did get something that looks like it wasn't totally wrong about the coin type - not a die match or a coin match. Seeing anything is then $45 - I think I will put that towards my next coin. Image search on ACSearch you can buy a year of all access for ~$120 and sure enough I found my actual coin (the first one from 2011). 4 Quote Link to comment Share on other sites More sharing options...
Prieure de Sion Posted July 19 · Member Author Share Posted July 19 I would like to emphasize that I do not want to ridicule the developer of the website. The idea itself is very good. Once a website like this works properly, it would be a fantastic thing! But the way I see it, there is not just a little work to be done here, but a really long way to go for this website and this tool. The current problem is that not even similar pieces are displayed as hits - instead, hits are displayed that have zero, really zero matches and represent a completely different motif. If I have a bust on one side and the search returns a modern coin with an eagle as a hit - or an ancient bust to the right returns a modern bust on a gold coin to the left. We're not talking about small differences and details that developers still need to get under control. I ask myself - if a Septimius Severus head returns a Campgate as a hit - how much more work is needed to make the search work in detail later on? Actually, the AI should not only deliver the same coins as the end product, but also differentiate that the source at 6 o'clock has a flan crack and the hits that do not have a flan crack are not displayed as hits. And until then, the developer still has a lot of work to do - even if the idea itself is of course very interesting - and would also be very helpful. 2 Quote Link to comment Share on other sites More sharing options...
Prieure de Sion Posted July 19 · Member Author Share Posted July 19 Ok 🙂 ... I'm a curious person and when I'm interested in something, I don't let it go so quickly. And especially here - when you criticize, you also have to report positive things when something works, otherwise it's just unconstructive bashing. So first of all, I tried another coin of mine, a Gordianus III sestertius. Unfortunately, the results were again not good. Then I remembered my Trajan bronze. The coin was “forcibly sold” under very sad circumstances in 1936 by a Jewish coin dealer in Frankfurt. If you want to read the whole story and research - you can find a thread about it in the German forum: https://www.numismatikforum.de/viewtopic.php?f=6&t=68304&hilit=Leo+Hamburger There is much suffering and human fate behind the coin and the time when Jewish businessmen were forced to give up their businesses. This coin was also sold in the last days of the Jewish coin store - in 1936. And up until then - 1936 - I was able to trace the coin, but not before. So I also searched for Trajan and actually found what I was looking for. The coin was offered in an auction by Leo Hamburger in 1932 - but probably not sold, as the coin was still in the store and sold in 1936 - or it was sold at the auction and later returned to the store. To have the provenance displayed, I had to pay 60 USD - but in the case of this coin it was worth it to me and I paid and got the provenance. So you actually “only” pay when you find a provenance and want to display it. However, I still have questions or constructive criticism for the developer. 1. after paying Stripe, I only received a confirmation for 60 USD. No invoice. Since the search should also be interesting, especially for coin dealers, a real invoice would be good so that you can deduct the amount from the tax! 2. if I had found this coin in three different catalogs - would I have had to pay 60 USD three times? I think so - right? That would be too much for me - you should pay for one coin and be able to see all the pedigrees for that coin. Otherwise, with 3-4 hits, it quickly becomes very expensive. 3. 60 USD is not a small amount - but I am prepared to pay it. However, I find it a pity that for this price there is only the picture of the auction catalog and only the picture of the plates with the coin - not the lot description! I will let the developer know these points (the developer is in the Ancient Coin Facebook group). PS: But my praise for the hit. As I said, if something is positive, it should be made public. Anyone can just criticize. 8 1 Quote Link to comment Share on other sites More sharing options...
Broucheion Posted July 22 · Member Share Posted July 22 Hi @Prieure de Sion, I’m sure you know rNumis shows description here: https://digi.ub.uni-heidelberg.de/diglit/hamburger1932_10_25/0041/image,info . But of course it should be included with the coin image when/if found. - Broucheion 1 1 Quote Link to comment Share on other sites More sharing options...
rasiel Posted July 22 · Member Share Posted July 22 This is a result of a very limited pool of training data. I'm starting work on something similar for Coryssa. Might end up being just as crappy as this but who knows. At least it won't cost anything. Wish me luck. Rasiel 3 1 2 Quote Link to comment Share on other sites More sharing options...
Nikodeimos Posted July 22 · Member Share Posted July 22 (edited) I actually had some success with the provenance tool. I'm mainly paying for the collection organizer but being able to make use of the provenance searcher as much as I want is really neat too. For this coin it found three provenances: Hirsch XII 1909 (Konsul Eduard Weber - already had that one), Helbing 1913 and Ars Classica XII (1926). For this coin it found two provenances: one to 1902 (the Hill catalog of the John Ward collection, already had that one - curiously, the provenance searcher calls it a 1970s pedigree, perhaps because it's using a later reprint) and Glendining & Empire Coins, 9 October 1989, 620. It may need some more work, but the potential is most definitely there. I also like how quick it actually goes. EDIT: It quite handily found the Münzhandlung Basel I 1934 pedigree for my Lucilla too Edited July 22 by Nikodeimos 3 1 Quote Link to comment Share on other sites More sharing options...
Prieure de Sion Posted July 22 · Member Author Share Posted July 22 8 hours ago, Broucheion said: Hi @Prieure de Sion, I’m sure you know rNumis shows description here: https://digi.ub.uni-heidelberg.de/diglit/hamburger1932_10_25/0041/image,info . But of course it should be included with the coin image when/if found. - Broucheion Thanks so much for the link - I downloaded the complete catalog. What great coins are offered in this auction 🙂 1 Quote Link to comment Share on other sites More sharing options...
Deinomenid Posted July 22 · Supporter Share Posted July 22 The related issue is simply that this whole process is being weaponised. The coincabinet one might not be great (hopefully just not yet) but there are strong reports of an exceptionally well-off US collector (not the one who is currently buying 12 examples of everything that was ever minted in the Greek world) who purchased practically all known catalogues before a certain date (including I believe a truly huge number from BCD at the time of but separate to the last auction there) who is systematically applying functioning artificial intelligence to them with very substantial resource behind it. In other words, soon there is likely to be at least one private collector who knows far more than the rest of us. The playing field is becoming more and more uneven, at least temporarily. Not complaining or tilting at rumoured windmills, just commenting on what seems to be a real change. I know little of what goes on outside my little Greek world (and little enough within that!) but I believe this individual has far broader numismatic interests. I suppose it is a logical progression of the vast wealth collecting in the hands of the technologically exceptional, and only needing one or two of them to develop this interest... 1 1 Quote Link to comment Share on other sites More sharing options...
rasiel Posted July 22 · Member Share Posted July 22 8 hours ago, Deinomenid said: The related issue is simply that this whole process is being weaponised. The coincabinet one might not be great (hopefully just not yet) but there are strong reports of an exceptionally well-off US collector (not the one who is currently buying 12 examples of everything that was ever minted in the Greek world) who purchased practically all known catalogues before a certain date (including I believe a truly huge number from BCD at the time of but separate to the last auction there) who is systematically applying functioning artificial intelligence to them with very substantial resource behind it. In other words, soon there is likely to be at least one private collector who knows far more than the rest of us. The playing field is becoming more and more uneven, at least temporarily. Not complaining or tilting at rumoured windmills, just commenting on what seems to be a real change. I know little of what goes on outside my little Greek world (and little enough within that!) but I believe this individual has far broader numismatic interests. I suppose it is a logical progression of the vast wealth collecting in the hands of the technologically exceptional, and only needing one or two of them to develop this interest... Well, that would be a waste of money if that's the intent because then there's guys like me who go and publish this info online for everyone to see. For free. If I wind up in a ditch somewhere... it wasn't suicide 🤭 Rasiel 2 1 Quote Link to comment Share on other sites More sharing options...
Deinomenid Posted July 23 · Supporter Share Posted July 23 (edited) If you have "all" catalogues pre whatever important recent date (r numis plus)and can automatically search them for images against a current auction's selection (the claim here) then I am hugely grateful for you leveling this playing field. Hugely! Please send me the link and I promise I will share it with all when I have supped my fill. I haven't seen anyone yet with anything remotely close. It would be like coincabinet fully functioning for image search and instead of manually inputting search coin by coin there, you can feed it hundreds and hundreds of images. I strongly suspect there's going to be a period of time when $ muscle combined with "AI" ability will make it worse for the likes of me and most of us, at least in terms of provenance search. I've seen an attempt made with the Lockett collection, or at least the photographed examples as there are plenty without, and it was getting impressive. Edited July 23 by Deinomenid typo Quote Link to comment Share on other sites More sharing options...
CPK Posted July 23 · Supporter Share Posted July 23 4 minutes ago, Deinomenid said: ...I strongly suspect there's going to be a period of time when $ muscle combined with "AI" ability will make it worse for the likes of me and most of us, at least in terms of provenance search... How does it make it worse exactly? It seems to me that having a working, comprehensive AI provenance search tool will make things better, not worse, even if you have to pay for it, like you would for most any other service. Having the option is better than not having it, surely? And anyway, whoever this is, I presume he's not buying literally every single printed catalog ever made? I mean, like all the copies printed? So anyone should still be able to purchase their own if they wish, for their own provenance research. Unless I am misunderstanding? 1 Quote Link to comment Share on other sites More sharing options...
Deinomenid Posted July 23 · Supporter Share Posted July 23 @CPKAhh sorry, yes that wasn't clear - not commercial and certainly not for free. I meant this appears to be for personal consumption, not for sale to you and me. As in an individual buys huge numbers of rare - eg BCD's catalogues, already done - plus the freely available ones such as the great selection on r numis and has a *functioning* coincabinet type program with an autofeeder of new auctions into that. It's a huge competitive edge, in my view, much as I truly appreciate the likes of coryssa et al. I'm absolutely not saying that over time this isn't replicable, but if one has an edge and $$$ there's a lot that can be bought until the rest of us catch up. 1 Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.