Shopping at LEGO or Amazon?
Please use our links: LEGO.comAmazon
As an Amazon Associate we earn from qualifying purchases.

LEGO bricks recognition app. Would it be useful?

radoscradosc Warsaw, PolandMember Posts: 3
Hi! First post here :) I came out of my dark ages just 6 months ago with an excuse of finding these cool star wars sets for my 7yo son. Got a LOT of used Lego lots from eBay. I think at this point we have like 40-50kg excluding 50 already assembled sets. 

Sorting Lego bricks was fun at the begging but now it is just painful. Researching each individual brick is pain in the ass. So since I run a startup that deals with new technologies I thought that it could be fun to build an app with brick recognition. Idea is to train an image recognition neural network. The way it would work is you'd need to snap a photo of an individual brick and it should give you a part number along with some less probable alternatives. It could than save it into your database, upload it to bricklink or make a list of sets that include this part. There's a lot of possibilities once recognition would be functional. 

Would you find it useful? I'd like to avoid hours of work if this wouldn't help anybody but me. 
Radek
SprinkleOtterdatsunrobbie

Comments

  • MrShinyAndNewMrShinyAndNew Member Posts: 281
    If it worked well it'd mean an end to the "Identify this piece" thread. 
    AllBrick
  • DaraghDaragh IrelandMember Posts: 363
    I think it would be incredibly useful to Bricklink sellers that sell used parts, I don't currently, but plan to in the future and one of the main reasons I don't is the searching for and locating parts. Will be very interested to hear how you get on with this project. 
    SprinkleOtterkiki180703
  • SprinkleOtterSprinkleOtter Member Posts: 2,751
    I would find that very useful.
  • Switchfoot55Switchfoot55 The Northwest, USAMember Posts: 2,228
    I'm not a major player in the "identifying parts" game, but I think the concept sounds awesome!
  • gmonkey76gmonkey76 ChicagoMember Posts: 1,672
    If it could aso have a list of all official colors that the brick was made in that would be very helpful.
  • AllBrickAllBrick UKMember Posts: 1,477
    If it worked well it'd mean an end to the "Identify this piece" thread. 
    That thread is like an app, the response to questions there can be incredibly fast.
  • ShibShib UKMember Posts: 5,421
    I think an app like this could be incredibly useful, but I'll be honest I really can't imagine it working. 

    First off, how good would the images the user feeds into it need to be? I know that most camera phones these days are pretty good, but being kept in pockets often leads to grubby lenses, also will lighting be an issue etc.

    The big question mark next to image quality for me is will it also be able to identify colour?

    How exactly will it identify parts? Counting studs? Not all parts have studs and will it be able to distinguish different part heights properly etc.

    Hav you had a good look through the Bricklink inventory to get an idea of just how many different parts have been made over the years? Don't forget that it's the more obscure parts that are hardest to identify too, so trying to trim down the amount of parts would likely make it far less useful. Also linked to this, where would you get your database of parts from? If you plan on using an existing database will you need permissions and will their images even be good enough? Again I suggest having a good look through some of the more obscure parts in the Bricklink inventory.

    Ultimately I do think it could be a very useful tool that could be made more useful by adding some of the additional data about parts (what sets its found in, other colours available etc) that a lot of LEGO sites already have. However I think you'd really struggle to get something working well, the market for it is relatively small and as most of the information on it could be found online you'd likely find most people unwilling to pay for the app.

    @AllBrick has hit the nail on the head there, the community generally does this pretty well without an automated app, pretty damn accurately and quickly.

    (p.s I don't want to be a total buzzkill about the idea, just being realistic, if you think you can make something good I'd be pleased for you to prove me wrong!)

    AllBrick
  • radoscradosc Warsaw, PolandMember Posts: 3
    I know how many  variants are there . The only reservations I have before starting this project is that 30k+ brick types can be a bit too much for simple neural network and I would probably have to somehow cluster it between a two. One to recognize type of a brick (like minifigs, technic, bricks, plates, etc.) and than pass it to other trained neural networks to recognise only these types. 

    Neural networks are a little bit like magic. You don't have to worry about individual components or how to recognize or count studs. You just feed it with brick images and it will learn. It'll train individual neurons to do that. Colors as well though there can be a problem with dark gray vs dark bluish gray because of camera white balance. And to clarify it can even be a crappy picture. It really doesn't matter as long as it's the only brick photographed. 

    I think I'll try that with something simple, like minifig torsos or something like that and an online interface for photo upload. 
    Daragh
  • datsunrobbiedatsunrobbie West Haven , CTMember Posts: 1,668
    The "what part is this" thread is awesome, and has the additional bonus that everybody on the forum gets to see the mysteries presented and solved. There have been multiple times that people have asked for help on that thread and posted parts that I was wondering about.

    That said, I think it sounds like a very cool app, and would come in very handy when searching thrift stores. Starting with minifigure torsos sounds great to me, as I have several hundred that I have not identified :) 
  • omniumomnium Brickenham, UKMember Posts: 831
    I think it would be a great app.  But I imagine the job of training the neural net with so many parts, from many angles, would be overwhelming.  You'd need a way of crowd-sourcing the training, I reckon.
  • radoscradosc Warsaw, PolandMember Posts: 3
    That is true. There are a few things involved here. First of all number of categories. Usually you'd supply 100 images of cat, 100 of dog and so on for a limited number of categories (like 20 or 30). Now for each brick, or even torso I'd have like 10-20 available if I'm lucky. Multiply that by 30k of categories. I'm not sure it'll work. So the idea is to start with something simpler like minifigs or minifig torsos. Should be less than 1k of categories but that's still a lot of work. In any case I think I'll train sample network and let you guys test it. And if successful add kind of crowd sourcing - a neural net that is preaty dumb but it would be retrained based on properly described photos from users. (if it fails, identify a brick yourself and input a number). 
  • ecmo47ecmo47 North CarolinaMember Posts: 2,101
    Most Lego parts have their 4 or 5 digit ID number printed on the underside. Might need a magnifying glass to see it, but it's usually there.
Sign In or Register to comment.

Shopping at LEGO.com or Amazon?

Please use our links: LEGO.com Amazon

Recent discussions Categories Privacy Policy Brickset.com

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Brickset.com is a participant in the Amazon Services LLC Associates Program, the Amazon.com.ca, Inc. Associates Program and the Amazon EU Associates Programme, which are affiliate advertising programs designed to provide a means for sites to earn advertising fees by advertising and linking to Amazon.

As an Amazon Associate we earn from qualifying purchases.