TLDR Summary

An extremely simple demo dataset with 5 classes.

Can be used to help learn how to train Darknet/YOLO, as it includes some annotations, trained weights, and several short videos: legogears_dataset.zip.

A new version of the dataset was released on 2024-05-30. This one has several new videos, annotations, and negative samples. This is the dataset that was used for this video which compared Darknet/YOLO and YOLOv10. Download it here: legogears_2_dataset.zip

Lego Gears

No copyright; the dataset is released as public domain.

For example, run this command:

DarkHelp --line 1 --autohide off LegoGears.cfg LegoGears.names LegoGears_best.weights DSCN1583A.MOV

(If not familiar with it, DarkHelp is a CLI tool as well as an API to access Darknet/YOLO. It runs on Windows, Linux, and Mac.)

See this YouTube video to see how this neural network was trained in less than 90 seconds!

Last modified: 2024-05-30
Stéphane Charette, stephanecharette@gmail.com