Deep Learning for Robotic Image Perception in GPUs: an Introductory Hands-on Tutorial

What is?

Deep Learning methods are currently the state-of-the-art in many image-related problems. This 6-hour tutorial explores Deep Learning for Robotics Image Perception in vector processors through a hands-on approach. Participants will have the opportunity to apply deep neural networks (DNNs) to image classification problems through a simplified set of tools, frameworks and data pipelines commonly used to train and deploy DNN in customised GPUs.

This is a hands-on tutorial. You should bring your own laptop to develop the GPU exercises during the course. Also, you should register to this tutorial by clicking in the button REGISTER above.


During the ICAR 2019 in Belo Horizonte, Brazil - Friday - 09:00-17:00 in the room 106. Please, BRING your OWN LAPTOP to the practical activities!"

ICAR 2019

19th International Conference on Advanced Robotics
The Program Click here


Undergraduate and graduate students, university lecturers and researchers interested in the field of deep learning for robotic image perception. Level: Elementary (70%) up to Intermediate (30%) Skills: Basics of programming. No previous experience is required in any of the machine learning tools and libraries explored in this tutorial: Caffe, TensorFlow, DIGITS e Python Jupyter Notebook.

Learning goals?

The expected outcome for each participant is that they gather an understanding of the fundamentals of deep learning in parallel processors (e.g. GPUs):
* To explore deep machine learning in vector processors in the cloud and embedded in real-world applications;
* To implement a pipeline of commonly used deep learning tasks;
* To experiment with datasets, fine-tune neural net training parameters, structure the inner layers of a network, and understand other strategies to improve the performance of a neural network;
*To integrate and deploy neural nets in the robotic field in order to solve real-world problems;
*To investigate the use of deep learning to deal with robotics challenges.

What is covered?

*Vector Processors and DSAs
*Introduction to Deep Machine Learning
*Deep Neural Networks - DNNs
*Performance and Accuracy of Models
*Recent Research Results in Deep Learning



Marcelo R. Pias

Professor at FURG
PhD in Computer Science - University College London - UCL


Paulo L. J. Drews-Jr

Professor at FURG
PhD in Computer Science - Universidade Federal de Minas Gerais - UFMG


Silvia S. C. Botelho

Professor at FURG
PhD in Robotics - Laboratoire d'Analyse et d'Architecture de Systèmes - LAAS/CNRS

Know More?

Tutorial in the SIBGRAPI 2019 - Conference on Graphics, Patterns and Images
Perfect Storm: DSAs Embrace Deep Learning for GPU-Based Computer Vision -Link