From pixels to predictions

One workflow, developed in stages

Image classification is a useful setting for learning the whole deep-learning workflow: prepare data, establish a baseline, train, evaluate, inspect mistakes, and improve the model.

From baseline to transfer learning

The Rock–Paper–Scissors capstone in my PyTorch course brings these stages together. Learners compare a baseline CNN with transfer learning and selective fine-tuning, then use the results to explain their decisions.

What learners practice

• Keeping training and evaluation data separate.
• Establishing a baseline before adding complexity.
• Inspecting class-level errors using a confusion matrix.
• Comparing models and saving a usable checkpoint.

The full notebooks, guided lessons, and completed solutions are part of the course. The free resources on this site provide standalone support for evaluation and experiment planning.