Lecture and Practical session on advanced CNNs

This material contains slides and a practical notebook on Convolutional Neural Networks (CNNs) for a practical session celebrated in the Ágora, our Journal Club from Universidad Carlos III de Madrid lead by the Signal Theory and Communications department. All material from the Ágora sessions including this one, can be found in this GitHub Repository. The session includes: A 30-minute theoretical overview of CNN building blocks (convolutions, activations, pooling, batch normalization), architectures (ResNet, U-Net, YOLO), transfer learning, fine-tuning, and evaluation metrics (accuracy, F1, mAP, Dice, IoU). A hands-on Jupyter Notebook designed to be run on Kaggle Notebooks, covering: Classification with custom CNNs and pretrained ResNet models. Object detection using YOLOv8. Segmentation using a custom U-Net and pretrained MONAI UNet. Loss functions and evaluation strategies tailored to each task. 📂 Datasets Used: Brain Cancer - MRI Dataset (Kaggle) Brain Tumor Detection Dataset (Ultralytics) Pancreatic Stem Cells Dataset (Cell Tracking Challenge)(Note: Dataset 2 and 3 must be manually downloaded and uploaded to Kaggle Datasets.) ⚠️ The notebook assumes basic familiarity with Kaggle Notebooks. If using private notebooks (required for internet access and downloading pretrained weights), account phone verification is needed on Kaggle. 🛠️ Tools and Frameworks: PyTorch torchvision MONAI Ultralytics YOLOv8 ✅ Citation: Please cite the original datasets, architectures, and tools if using this notebook for research or educational purposes. Dataset citations are included within the notebook.