Free online courseConvolutional Neural Networks for Computer Vision: Detection, Recognition

Course Description

Want to teach machines to see and understand images? This free online course dives into Convolutional Neural Networks (CNNs), the foundation behind many modern computer vision systems used for object detection, image recognition, face verification, and even neural style transfer. You will build intuition for why convolutions scale so well to large images, how filters capture visual patterns such as edges and textures, and how design choices like padding, stride, pooling, and volumetric convolutions shape what a network can learn.

As you progress, you will connect the core building blocks to practical, battle-tested architectures. You will explore why classic network designs matter, what made very deep networks finally trainable, and how ideas like skip connections, bottleneck layers, and multi-branch modules help models become both stronger and more efficient. Along the way, the course emphasizes the kind of reasoning used in real projects: choosing proven baselines, adapting open-source implementations responsibly, and applying transfer learning to reach high performance even when data or training time is limited. You will also see how data augmentation can improve robustness by exposing a model to the variety it will face in the real world.

From there, the focus shifts from recognizing a whole image to pinpointing what is inside it. You will learn how localization augments classification with additional outputs, how detection systems evaluate predictions using IoU, and why post-processing steps like non-max suppression are essential for turning dense candidate predictions into clean final boxes. Modern detection approaches are covered conceptually, helping you understand the key tradeoffs between sliding-window style computation, region proposal pipelines, and single-shot methods such as YOLO.

The final part broadens your toolkit with techniques for identity and similarity learning, including one-shot learning for face recognition using Siamese networks and triplet loss. You will also look inside CNN representations to understand what different layers tend to respond to, then apply that insight to neural style transfer, where optimization balances content and style statistics to generate new images. By the end, you will be able to reason clearly about CNN design choices, evaluation signals, and the components that power detection and recognition systems—skills that transfer directly to many AI and machine learning roles in technology.