awesome-lego-machine-learning

A short talk with Brian Schwab without many details about the augmented reality for playing with Lego bricks.

Podcast (and transcript) about LEGO data platform.

Description of the winning project of the internal Generative AI Hackathon at the LEGO Group. It focuses on generating a set description based on metadata and an image.

Podcast about using ML at LEGO. In particular, their ML infrastructure, content moderation system, and classifying images of LEGO models into thematical categories.

Short description of internal Data Science hackathon.

It introduces an automatic system to convert an architectural model into a LEGO sculpture while preserving the original model’s shape features like repeating components, shape details and planarity.

Generating sets step by step from the 2D images using RL.

Almost 500 original and custom sets.

Given the input portrait, it uses several deep neural networks to extract the attributes to describe the human appearance shown in the image. Built on these attributes, the model generates the corresponding uncolored brick model by iteratively searching for brick components with the coordinate descent algorithm. Finally, it assigns a color to every brick to get the final brick model.

The system takes an image as an input, encodes it as embedding, generates a voxelized 3D model, and finally convert it into LEGO bricks. The model jointly trains a 3D model autoencoder and a 2D-to-3D encoder.

LDraw repository of over 1800 official LEGO sets.

Students' report containing the design of the sorting machine. Mostly related to the mechanical side of the machine.

Dataset for parts classification. It contains 800k high-quality renders for 200 different parts.

Dataset for parts detection. It contains 2k high-quality renders for 200 different parts.

Initiative to build Lego AI Open Source Sorting Machine.

Mobile application which detects up to 2k different parts.

The paper describes the process of using synthetic data for training semantic segmentation model for detecting Lego parts.

General list of LEGO resources.

Introduction of a new task for visual understanding, 3D simulator to manipulate LEGO models, and a model to solve the proposed task of recreating the LEGO model.

It proposes a way to represent the LEGO model as graphs and learn how to generate them step-by-step.

A graph transformer-based framework for ASP problem, with a heterogeneous graph attention network to encoder the models, which are decoded with the attention mechanism to generate assembly sequence.

This paper presents a system that automates LEGO assembly and disassembly by learning from human demonstration.

Understanding the assembly process using manual.

Generating LEGO Pirates sets and minifigures using Stable Diffusion.

Stable Diffusion model for generating Brickheadz.

Stable Diffusion model for generating Minifigures.

Fine-tuned stable diffusion model for generating images in the LEGO style.

Generating realistic versions of LEGO sets.

The post discusses how AI-generated “LEGO” sets are sparking a debate between inspiration and imitation within the AFOL community.

Using machine learning algorithms to predict future prices of sets.

A dataset of 153 Lego sets focusing on detailed 3D object reconstruction with a unique data modality of high-resolution RGB images with low-resolution depth maps captured on a mobile device.