There aren't that many open source libraries for computer vision, as far as I know. Among the few, Luminoth is one of them. Built in Python, it uses two large libraries: TensorFlow and Sonnet. It supports Python 2.7 and 3.4 to 3.6.
It can be installed with either CPU or GPU support.
For CPU you install TensorFlow and Sonnet first, then you run: pip install luminoth.
For GPU you install both TensorFlow and Sonnet with GPU support and then you run: pip install luminoth.
As of now, Luminoth supports image classification and object detection. However, they plan to add support for YOLO, SSD, and Mask R-CNN. They also want to provide pre-trained models on popular datasets.
If you're just starting out and you want to play with some machine vision you could simply use their framework and I'd suspect you'd be doing the CPU version of it. There's not much 'deep' work that you can do on the CPU alone, but I'd say that's enough to get you buzzed about the possibilities.
Additionally, you could also play with machine vision on IBM Watson and also by using the classify_image.py from the official TensorFlow - but these are very automated and you don't get to build the entire thing yourself (hence it may limit your learning).
You can read more about Luminoth and you can also follow their official (short) tutorial on the following github repo:
@cristiCristi Vlad Self-Experimenter and Author
