There is a decent argument that computer vision is now the least fashionable part of applied AI, which is odd given how much of the work in manufacturing, medicine, agriculture and robotics is exactly that. OpenCV University exists because the organisation behind the most widely used vision library decided to teach the subject properly, and the result is the most focused training in this niche. The catalogue runs from Mastering OpenCV with Python, a two to four week entry course, through Fundamentals of Computer Vision and Image Processing at about three months, into the longer deep learning tracks in either PyTorch or TensorFlow and Keras at four to five months each, plus an advanced course on vision applications with deep learning and transformers, and a lighter offering on generative AI for art. These are also sold as stacked programmes under the CVDL banner.
The strongest thing here, and the reason I would recommend it over a generic deep learning course for anyone serious about vision, is the classical material. Most modern curricula skip straight to convolutional networks and leave students unable to handle the enormous number of vision problems that a threshold, a morphological operation or a well chosen colour space solves in ten lines. Understanding what happens to an image before a model ever sees it is what separates people who can ship a working vision system from people who can fine tune a model and then cannot work out why it fails in the client's warehouse lighting. OpenCV University teaches that layer properly and the assignments reflect it, using awkward real images rather than the clean benchmark sets that make everything look easy.
The support model is also better than most self paced products. You get lifetime access to the material and 24 months of forum support, and the forums are genuinely monitored by people who know the library, which matters when you are stuck on something that is a quirk of the API rather than a gap in your understanding. Three honest criticisms. First, the pricing presentation.
Almost everything in the catalogue is permanently shown as discounted from a higher number, and after a while the original prices stop meaning anything. I would rather they simply charged the real price. Second, the bundles get expensive fast. Paying $2,499 for the full stack of self paced video and assignments is bootcamp money without the cohort, the mentor or the careers support, and I would steer nearly everyone towards a single course to start, since they stand alone and you can add later without losing anything.
Third, and this is a scoping point rather than a fault, the education is narrow by design. You will come out able to do computer vision and not much else. If you want a general foundation in machine learning, start elsewhere and come here when vision is definitely the direction. The other gap worth flagging is deployment.
The model building content is strong, but the material on getting a vision system running reliably in production, on edge hardware, under variable conditions, is thinner than I would like given how much of the real difficulty lives there. My 4.1 reflects genuinely good, well targeted teaching from the most credible source in the field, held back a little by pricing theatrics and by the deployment gap. If vision is your thing, start with the fundamentals course and decide from there.