2024-10-26 19:25:00
rebrickable.com
How does RebrickNet work?
RebrickNet learns by looking at images of LEGO parts and discovering the features that make each part unique.
AIs like RebrickNet use machine learning
to discover relevant features. This learning process requires millions of images under a wide variety of
lighting conditions.
We can accept two types of submissions for training data:
- (Preferred) Images of a pile of LEGO parts, all of a single type. This makes labelling very easy, provides a better quality
source image for training, but requires more submissions to get the same quantity as a video. - Videos of rotating LEGO parts. A single 10 sec video can contain ~50 usable frames, making it an efficient way to get
the images needed. However, it requires significant processing prior to training, and the final result is not as
high quality as a photo of multiple parts taken together.
Login to submit training videos/images.
How to create RebrickNet images for submission?
Submitting images is far easier than videos. Simply grab a bunch (~ 10 to 20) of a single part type and place them
on a flat surface. You can take many photos of the parts (aim for at least 10), shuffling them in between each time.
The background can be anything as long as it doesn’t contain LEGO images or other parts. Using a variety of
backgrounds helps too.
How to create RebrickNet videos for submission?
What LEGO Parts do we need?
We are starting simple and slowly growing in complexity. To achieve this, we are splitting the parts into batches.
Each batch represents a new milestone in how many LEGO parts RebrickNet can predict. You can submit any of the parts
below (including any mold variations), from any batch. But please look at the RebrickNet Submission Stats
page to determine which parts are in more need of submissions than others.
Watch this forum topic to be notified of the
changes to requirements. As RebrickNet evolves, we will change what parts/colors are needed and will only be
accepting uploads of the ones we need. This is to keep this project at a manageable level and
keep costs to a minimum.
Login to use RebrickNet.
Release: v0.268.130.55.7 (2022-10-20)
•
Parts: 300
•
Colors: 64
•
Parts Accuracy: 93%
•
Color Accuracy: 80%
•
Combined Accuracy: 77%
•
False Positives: 1%
Support Techcratic
If you find value in Techcratic’s insights and articles, consider supporting us with Bitcoin. Your support helps me, as a solo operator, continue delivering high-quality content while managing all the technical aspects, from server maintenance to blog writing, future updates, and improvements. Support Innovation! Thank you.
Bitcoin Address:
bc1qlszw7elx2qahjwvaryh0tkgg8y68enw30gpvge
Please verify this address before sending funds.
Bitcoin QR Code
Simply scan the QR code below to support Techcratic.
Please read the Privacy and Security Disclaimer on how Techcratic handles your support.
Disclaimer: As an Amazon Associate, Techcratic may earn from qualifying purchases.