Module 2 · Project 1: Facial Emotion Detection (image classification with CNN)

OBJECTIVE: build an app that looks at a face (webcam or photo) and tells the emotion — angry, disgust, fear, happy, sad, surprise or neutral — and learn CNNs and image data while building it. WHAT WE COVER (in project order): • Images as data: pixels, channels, size • Image data preparation: collecting, labelling into folders, resizing, splitting, augmentation, class imbalance • CNN concepts: convolution, filters, pooling, CNN architecture • Building and training a CNN in Keras • Transfer learning with a pretrained model • Evaluation: accuracy, per-emotion recall, confusion matrix, Grad-CAM • Deployment: face detection + emotion model in a Gradio app AFTER THIS MODULE YOU CAN: • Prepare any image classification dataset end to end • Build, train and improve a CNN, and use transfer learning • Evaluate an image model honestly and explain its mistakes • Ship an image model as a public web app DATASET: FER-2013 (about 35,000 face images, 48×48 grayscale, 7 emotions) + a few of your own photos for a real-world test. TOOL: TensorFlow/Keras.