Image Recognition Watch Accuracy App. Any interest?

[3rdpirate]

Greetings ,

Hope all are well and safe.

I've been taking an interest in ML and specifically image recognition algorithms for the last several years as a side hobby. I have joined a BOSCH research group at the university I formerly graduated, dedicated to this topic.

Would any of you be interested in an app that works on the following principle :

1. Open app, take a shot of your watch.
2. Watch snippet will be timestamped according the internal devices clock.
3. App will analyze the snippet for H/S markers and tell you the time the identified on your watch and ask you for the confirmation. (might not be accurate in the first phases due to large amount of data needed)
4. WAIT as much as you consider appropriate.
5. Take a shot of the watch again.
6. The app will identify the H/S markers and ask for confirmation. Then compute the actual delay of your watch's movement in the time frame between the shots. Alternatively it can estimate the accuracy on a larger time frame given, be it a day, a month or a year.

This will probably be done regardless of your interest, but if there's any, I can make it a public repository and create a pleasant UI/UX.

Rant over.

Take care! Any input welcome.

 

[Teetertank]

I was wondering just the other day if this existed somewhere. Never seen it in an app before but figured it would be a good idea.

I use WatchCheck on Android, but obviously need to manually press the measure button. 'Watch Accuracy Meter' is also good for making instant measurements and adjustments. Both apps have their place - WatchCheck for long-term daily wear rate check and Watch Accuracy Meter for making adjustments based on results from WatchCheck figures.

So, did you make any progress on this?

Probably wouldn't be too difficult on plain faced watches, but others with complications and strange date windows would cause more of an issue for the image recognition, though these days could well be a piece of cake given the advancements in AI.

 

[attachedthanks]

Did you ever end up making this? Cool concept. I have been learning ML myself lately and love it!

 

[stagename42]

That would be really interesting, although probably hard to implement for watches that have rare or custom features which is typically what people are trying to find online when they see it somewhere