Watch movement recognition software

[HectorLooi]

I was wondering if it is possible to create an app that can identify an unknown movement just by taking a photo of it.

There are plant identification apps available at this present moment. I think the algorithms should be similar.

So all the software designers out there, what say you?

[Poljot]

3 hours ago, HectorLooi said:

I was wondering if it is possible to create an app that can identify an unknown movement just by taking a photo of it.

There are plant identification apps available at this present moment. I think the algorithms should be similar.

So all the software designers out there, what say you?

You mean an app when people forget how to read?

[Nucejoe]

No expert here. Sure it is possible.

I tried googling some pictures of watch movements, it either bring me the same picture or of some other watches and occasionaly pots and pans instead, oh and an arse once by the name of Rolex.

Who else would have the required archive of pictures. 

 

 

[wls1971]

It would be a very useful app "Bestfit" and "Paulsons master key" are by far the best way of identifying watch movements because as anyone who has dealt with watches form the early 1900-1940 identification can be the biggest hinderance to finding parts, identifying the keyless works is the only way because we all know that a lot of  movements from that period do not have nice shiny calibre numbers engraved in prominent places to be read.

Edited December 5, 2020 by wls1971

[KOwatch]

Ranfft has an OK advanced search feature, but it's very manual and leaves a lot to be desired. With a bit of practice, it'll give you a good list of candidates which will  include your answer or at least a similar movement from the same design, if it's in his database. You can limit your search based on the bridge layout details and some features. You can't, for example, limit your search by size or by approximate year even though this info is in the db. Sometimes this can be an advantage, if for example you've got an unspecified model that is based on an earlier or different sized design. It also seems to not work well if you input too many parameters. 

If you could take his database and improve the search querying and the sorting, plus add some image recognition to the movement ID scheme, it would be fun to try.

 

 

[oldhippy]

There are countless movements that get posted on here for I D many that don't have any markings at all, the only way is by looking at the keyless  work or going on line, not many on here have old reference books. I posted my Bestfit catalogues some time ago on here for people to download but even so they don't have everything. Its a must, so who ever you are make it happen.  

[Nucejoe]

Did you have the app to be based in WRT? 

Good idea.

[AshF]

It's almost impossible (currently) to make an app like this.  The plant finder apps or similar use "convolution neural networks" that are trained on thousands of user identified pictures of the subject.  They're particularly good at this task as the subject has a fractal nature to it.  Watch movements however aren't.  The subtle differences between the size and shape of the keyless works would be hard to match to training images.

If I were tasked to program this I'd use an "expert system".  It would require more work from the user, they'd have to measure parts etc but it would be more accurate in finding the correct movement.

For example, the app might start by asking you "Does the movement have sweep seconds", and you'd answer yes, no or unknown.  From that it could half the movements it knows about a bit like playing 20 questions.

[khang]

That is a good software idea especially for those amateurs like myself. On my half, if someone decides to develop such an app, I sure will take high-res photos of my watch movements in order to contribute (almost all are vintage Citizens).