Use this data for training custom LEGO object classification models. This highly realistic data is fully synthetic, and attempts to mimic photo-realism as closely as possible.
These images are in context, meaning objects are set in environments with random floor textures, random lighting conditions, and each image has the possibility of containing other LEGO parts, with occasional obstruction. This creates the opportunity to train an extremely robust model that is exposed to realistic expectations.
✔️ 200 Most Popular LEGO Parts
✔️ 4,000 Images Per LEGO Part
✔️ 800,000 Total Images
✔️ 64x64 RGB Images
✔️ In Context Images
This data was created through a mixed usage of the Blender Python API alongside many other Python packages including Matplotlib, Pillow, and PyAutoGUI.
Brick Architect (https://brickarchitect.com) for knowledge and resources on LEGO parts and colors. LDraw (https://www.ldraw.org/) for 3D part models.