RebrickNet
databases & reference, software & apps
RebrickNet is Rebrickable's photo-based part identifier, matching an uploaded picture of a LEGO piece against its part database via a trained neural network.
ai parts of this page were drafted with ai; publication approved by a human — more
- pricing
- Free
- platforms
- Web
- status
- Active
- public api
- yes
- rlfm status
- Unofficial
- content format
- Web
- author
- Rebrickable Team
- update frequency
- Monthly
Overview
RebrickNet is a computer-vision tool built by the Rebrickable team: upload a photo of a single LEGO element and it returns likely matches from Rebrickable's part catalog, ranked by the network's confidence. It's aimed at identifying unlabeled or unfamiliar parts (inherited lots, vintage pieces, or molds a builder doesn't recognize) rather than sorting a full pile at once.
The model improves from user feedback: confirming or correcting an identification feeds back into training data, and recognition is generally stronger for common, distinctly-shaped parts than for close variants like similar hinge or bracket revisions.
Goals
- Automate the identification of LEGO parts using machine learning.
- Integrate part identification directly into the Rebrickable ecosystem and inventory management.
- Provide a scalable solution for high-volume part sorting and cataloging.
Characteristics
Neural Network Integration
Trained on millions of images to recognize 3D shapes.
Photo-to-Part Matching
Direct conversion of visual data into Rebrickable part IDs.
Continuous Learning
Accuracy increases over time through community feedback and new training data.
Pros
- Built for identifying a single unfamiliar part from a photo, not sorting a full bulk pile at once.
- Improves from user-confirmed corrections feeding back into the training data over time.
Cons
- Works one part at a time from a clear photo, not a bulk pile scan.
- Recognition is weaker on near-identical mold revisions, like close hinge or bracket variants, than on distinct shapes.