Introduction to computer vision with TensorFlow: Review

Even for a Microsoft Learn module, this one is short. The core of it is clicking through seven sections that are really just web pages with code snippets. It felt more like reading a well-illustrated blog post than taking a course. You don't actually run anything. The code is there, showing you how to load MNIST or build a simple CNN in TensorFlow, but you're not in a notebook, not debugging, not even typing. You just read the block and click ‘Continue’.

I finished it in about 45 minutes. The explanations for convolutions and transfer learning were actually quite good for being so brief. They use diagrams to show how filters slide over an image and explain why you'd use a pre-trained model like VGG16 instead of starting from scratch. That part was helpful for connecting the dots from my theory lectures. But the 'Train a simple dense neural network' part was just a few lines of Keras. It shows you model.fit(), and that's it. There’s no discussion of what happens if it doesn't converge, or how to tune hyperparameters.

The final 'Knowledge check' quiz was also disappointing. The questions were mostly vocabulary checks from the text you just read. I got 100% without really thinking, which makes me question how much I actually learned versus how much I just memorized for two minutes. It's a decent map of the territory, but it's all map and no walking.