A hands-on tutorial which will allow you to practice and finally train cutting-edge Deep Learning models, covering the most active and demanded skills in the field: Image Classification, Feature detection and Image Segmentation.
This tutorial will allow you to grasp of the fundamental concepts you need to solve common Computer Vision problems (Classification, Detection, and Segmentation), using state of the art Deep Neural Models, with the help of two of the most well known Machine Learning libraries, Keras and Tensorflow.
This talk is ideal for Python Programmers who want to have a quick and practical introduction to Computer Vision, and to be able to apply these concepts solving their own problems
Pre-requisites: The participants should have a basic/intermediate knowledge of the Python language, and the main Machine Learning concepts. The training itself requires bringing a notebook Jupyter Notebook along with the Keras library installed. Some of the tasks are computationally expensive, so a powerful notebook will reduce the training time greatly. Having the last version of Anaconda installed, with keras and opencv additional packages installed via "conda install keras opencv" will be definitely a plus.
Strategy: Jupyter notebooks and sample datasets will be provided to allow a quick setup and running of the initial tasks, and at the same time allowing for free experimentation and enrichening of the models, as the training progresses.
Outline
Intro and environment setup
A tour of Neural Models with Fashion MNIST: In this section we will perform clothing classification, using a variety of models, starting from a simple Perceptron, to very advanced Convolutional Neural Networks.
Using the best parts (Transfer Learning for classification) In this section, we will use the visual knowledge stored in a state of the art Deep Neural Network, adapting it to recognize visual classes of our interest.
Detect your own stuff (Object detection with Tensorflow Object Detection API): In this section, we will train an advanced lightweight Deep Neural Network, to localize objects of our interest.
Separate what's distinct (Object Segmentation): This section will allow you to detect and select elements of interest in your images.
Final Topics and conclusion: In this section, we review other existing techniques, and will review how the future looks for the field of Computer Vision.