/courses/machine-learning/computer-vision-cnns/Recent content in Computer Vision with CNNs on DATATWEETSHugoenCopyright (c) 2025 DatatweetsFri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/computer-vision-cnns/lesson-1-introduction-to-convolutional-neural-networks/Fri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/computer-vision-cnns/lesson-1-introduction-to-convolutional-neural-networks/Learn why dense networks struggle with images and how convolutions solve the problem. You will explore local receptive fields, kernels, feature maps, parameter sharing, and translation invariance, then meet Fashion-MNIST, the clothing dataset you will use throughout this module./courses/machine-learning/computer-vision-cnns/lesson-2-cnn-architecture/Fri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/computer-vision-cnns/lesson-2-cnn-architecture/Learn how the core building blocks of a convolutional neural network fit together: Conv2D, MaxPooling2D, Flatten, Dense, and a softmax output. You will stack them into a keras.Sequential model, compile it, and train a baseline CNN on the real Fashion-MNIST dataset, reaching 88 percent test accuracy and spotting the train/validation gap that motivates the next lesson./courses/machine-learning/computer-vision-cnns/lesson-3-regularization-in-deep-learning/Fri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/computer-vision-cnns/lesson-3-regularization-in-deep-learning/Learn why deep CNNs overfit and how to fix it. You will diagnose a strong but overfitting baseline on Fashion-MNIST, then add dropout and data augmentation to close the train/validation gap, and understand why a slightly lower test accuracy with a smaller gap often generalizes better./courses/machine-learning/computer-vision-cnns/lesson-4-advanced-cnn-architectures/Fri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/computer-vision-cnns/lesson-4-advanced-cnn-architectures/Move beyond a basic CNN. You will build a deeper, batch-normalized network on Fashion-MNIST, learn what global average pooling does, and meet the ideas behind classic architectures like VGG-style blocks and residual connections. Along the way you will see why deeper is not automatically better on a small dataset./courses/machine-learning/computer-vision-cnns/lesson-5-transfer-learning/Fri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/computer-vision-cnns/lesson-5-transfer-learning/Learn how transfer learning lets you borrow the visual knowledge inside a model pretrained on millions of images. You will load MobileNetV2 with ImageNet weights, freeze it as a feature extractor, build a small head, and reach near-baseline accuracy on Fashion-MNIST using only a fraction of the data./courses/machine-learning/computer-vision-cnns/lesson-6-guided-project-detecting-pneumonia-from-x-ray-images/Fri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/computer-vision-cnns/lesson-6-guided-project-detecting-pneumonia-from-x-ray-images/Bring your CNN skills together in a guided medical-imaging project. You will load the real PneumoniaMNIST chest X-ray dataset, build and train a small CNN with dropout, evaluate it with a confusion matrix and AUC, and learn why recall is the metric that matters most when missing a diagnosis can cost a life.