Stanford University
The definitive computer vision deep learning course, with public notes and assignments on CNNs, training tricks, detection, segmentation, generative models, and vision transformers. Self-study it when you want to build vision systems from scratch rather than only call APIs.
Developers with deep learning basics. This is an advanced course in Computer Vision, running about 10 weeks (one quarter).
Free. Current-quarter recordings are Stanford-only; prior-year lectures are on YouTube. Notes and assignments are public.
Implements object detection, image segmentation, and model interpretability (class activation maps, saliency) in TensorFlow. For TensorFlow users who already know CNN basics and want to ship detection and segmentation models.
Build an image classifier with convolutional networks and transfer learning in Keras, with notebooks you run in the browser. A compact starting point for Kaggle users who want to enter their first vision competition.
Covers image basics, OpenCV and Pillow processing, classical classifiers, CNNs, and object detection, ending with a deployed web app project. Suited to developers who want applied vision skills without the graduate-level math of CS231n.