Luminoth is an open-source computer vision toolkit from Tryolabs that packages state-of-the-art object detection models behind a single command-line interface and Python API. Released under the BSD-3-Clause license, it gathered more than 2,400 GitHub stars; the codebase was last updated in 2023 and stands as a complete, if no longer actively developed, deep learning toolkit.
Built on TensorFlow, Luminoth implements the Faster R-CNN architecture end to end โ dataset handling, training loops, evaluation and inference โ so practitioners do not have to stitch together research code. It ships with pretrained checkpoints, configurable training pipelines and utilities for preparing datasets in common formats. The project grew out of Tryolabs’ client work in applied computer vision and reflects that production orientation: sensible defaults, checkpoints management and a web-ready serving path rather than notebook-only experiments.
Luminoth is free and open source under the BSD-3-Clause license, with no paid edition or hosted service attached.
Luminoth fits machine learning engineers and students who want a readable, self-contained TensorFlow codebase for learning or prototyping object detection, and teams maintaining legacy detection pipelines built on it.
In practice, Luminoth is driven through its lum command-line tool: prepare a dataset, adjust a YAML configuration, then train, evaluate and predict with short commands. It also ships a small serving mode for exposing a trained detector over HTTP, and its documentation walks through the full workflow on standard benchmarks. Although newer frameworks have since taken over most object detection work, Luminoth remains a clear, complete reference for how a Faster R-CNN pipeline fits together end to end.