Advanced Computer Vision with TensorFlow

Overview

Class Central Tips

In this course, you will: a) Explore image classification, image segmentation, object localization, and object detection. Apply transfer learning to object localization and detection. b) Apply object detection models such as regional-CNN and ResNet-50, customize existing models, and build your own models to detect, localize, and label your own rubber duck images. c) Implement image segmentation using variations of the fully convolutional network (FCN) including U-Net and d) Mask-RCNN to identify and detect numbers, pets, zombies, and more. d) Identify which parts of an image are being used by your model to make its predictions using class activation maps and saliency maps and apply these ML interpretation methods to inspect and improve the design of a famous network, AlexNet. The DeepLearning.AI TensorFlow: Advanced Techniques Specialization introduces the features of TensorFlow that provide learners with more control over their model architecture and tools that help them create and train advanced ML models. This Specialization is for early and mid-career software and machine learning engineers with a foundational understanding of TensorFlow who are looking to expand their knowledge and skill set by learning advanced TensorFlow features to build powerful models.

Syllabus

Introduction to Computer Vision

Get a conceptual overview of image classification, object localization, object detection, and image segmentation. Also be able to describe multi-label classification, and distinguish between semantic segmentation and instance segmentation. In the rest of this course, you will apply TensorFlow to build object detection and image segmentation models.

Object Detection

This week, you’ll get an overview of some popular object detection models, such as regional-CNN and ResNet-50. You’ll use object detection models that you’ll retrieve from TensorFlow Hub, download your own models and configure them for training, and also build your own models for object detection. By using transfer learning, you will train a model to detect and localize rubber duckies using just five training examples. You’ll also get to manually label your own rubber ducky images!

Image Segmentation

This week is all about image segmentation using variations of the fully convolutional neural network. With these networks, you can assign class labels to each pixel, and perform much more detailed identification of objects compared to bounding boxes. You’ll build the fully convolutional neural network, U-Net, and Mask R-CNN this week to identify and detect numbers, pets, and even zombies!

Visualization and Interpretability

This week, you’ll learn about the importance of model interpretability, which is the understanding of how your model arrives at its decisions. You’ll also implement class activation maps, saliency maps, and gradient-weighted class activation maps to identify which parts of an image are being used by your model to make its predictions. You’ll also see an example of how visualizing a model’s intermediate layer activations can help to improve the design of a famous network, AlexNet.