← Computer Vision learning pathComputer Vision · Module 4 of 10
Image classification
Label whole images with deep learning.
📖 Key ideas
- Classification predicts one label per image (cat, dog).
- CNNs learn features from pixels.
- Datasets: MNIST, CIFAR-10, ImageNet.
- Transfer learning with ResNet/EfficientNet works with small datasets.
- Metrics: accuracy, confusion matrix, top-5 accuracy.
💡 Real-life example: Image classification is a librarian putting each photo into the right labelled box.
ClassificationAssigning a label to an image
ImageNetLarge labelled image dataset
Confusion matrixTable of predicted vs actual labels
Top-5 accuracyCorrect label among top 5 guesses