News for hackers. Stuff that matters.
Re:Rebrickable – Build with Lego by eranation on Sunday November 22, 2020 @01:26PM (#25179657)
Did you or anyone you know work on the obvious issue of cataloging the vast piles of legos out kids have using computer vision / AI? Cataloging inventory by hand is so discouraging. Ps. I didn’t check but I assume I can just upload some CSV with what parts I have etc? The issue is that doing it by hand sounds like a nightmare. Any ideas of attempts to auto scan parts on the floor and giving some estimate at least using open CV etc? (I should Google it first before asking probably...)

Edit: are solutions like this practical, or it’s easier to just sort by hand?

https://towardsdatascience.com/machine-learning-lego-image-r...

Re:Rebrickable – Build with Lego by bhaak on Sunday November 22, 2020 @03:47PM (#25180915)
If your question is about getting the contents of a Lego set, it is easier than it used to be.

First, every building instruction manual these days has the full inventory of pieces on the last pages.

Second, you can download a PDF of the instruction without owning the set. For example on page 43 on this one https://www.lego.com/en-us/service/buildinginstructions/7519...

Re:Rebrickable – Build with Lego by clickok on Sunday November 22, 2020 @03:52PM (#25180977)
Depends what you mean by "practical". Assuming you've done CV/ML stuff before, you could probably get something working pretty well over a weekend, and I think I could solve it completely with a bit more effort via 3D scanning + synthetic dataset generation... but unless you have cubic meters of lego to sort, doing it by hand would be faster, albeit less fun.
Re:Rebrickable – Build with Lego by someotheridiot on Sunday November 22, 2020 @04:35PM (#25181329)
There have been several attempts that I know of. But they all tend to have a low number of categories they sort into - tens vs the thousands of actual categories that exist.

I have tried it myself and it simply needs more input data (ie photos). However, that is something I might follow up on later as I can fairly easily get more data just by asking my community :)