ICIP 2016 is pleased to offer multiple workshops hosted by our patrons.

Imaging/Computer Vision Technologies at Intel Corporation

Intel is a world leader in silicon innovation and one of the largest manufacturers of computing devices. Intel conducts multi-disciplinary research for computing innovations. As part of Intel’s commitment to bring innovative computer vision and image processing technologies to the research community, Intel is a platinum sponsor of the IEEE International Conference on Image Processing (ICIP) 2016. In addition, Intel is organizing a workshop session and a demonstration booth to showcase the technology leadership. The details of the workshop session and demo booth are given below. For further information please contact: mahesh.subedar@intel.com or mahesh.subedar@intel.com

Intel Demo Booth

Booth #11
Intel is organizing a demo booth to bring cutting edge technologies to the research community at ICIP 2016. The demonstrations will showcase compelling usages and tools which will facilitate the researchers to benefit from the Intel architecture and the software tools.

Intel Workshop

The aim of the Intel workshop is to connect with the research community and provide information on imaging and computer vision technologies developed on Intel architecture. In this workshop we will share the ongoing research work at Intel, along with the architectural advances in the upcoming products. The details of the SW tools and SDKs will be shared which can benefit the research community in conducting advanced research in these areas.

Workshop Presentations

1) Interactive multimodal scene understanding
Presenters: Dr. Mahesh Subedar, Research Scientist, Intel Labs.
Dr. Omesh Tickoo, Computer Vision Research Manager, Intel Labs.
Description: The talk will introduce the efforts to portably and efficiently combine multi-sensor information for understanding the static and dynamic environments around different agents. The interactive portion of the scene understanding focuses on efficient descriptions for machine representation of the scene.

2) Bringing Enhanced 4K Experience in latest Graphics Architecture
Presenter: Dr. Wen-fu Kao, Media Architect, Visual and Parallel Computing Group (VPG).
Description: The talk will introduce new media features and the optimizations in the latest SkyLake and KabyLake Graphics Architecture to bring the best end-to-end 4K media experience to the latest PC and mobile devices.

3) Heterogeneous computing for image processing
Presenter: Dr. Kari Pulli, Sr. Principal Engineer, Imaging and Camera Technology Group (ICG), Intel
Description: Modern SOC’s (systems-on-a-chip) contain several processors, such as a CPU, GPU, and an IPU. In this talk we discuss what is an IPU, an Imaging Processing Unit, and various tools that facilitate harnessing the computing power of these processors.

4) Computer Vision SDK and its capabilities
Presenter: Jeffrey Mcallister, Technical consulting engineer, Software Solutions Group (SSG)
Description: Based on OpenVX*, the Intel® Computer Vision SDK lets you harness the performance of computer vision accelerators from Intel in your processing pipeline. In this talk, we show you how to create your own workload graphs with the included Vision Algorithm Designer to create a complete, functional application using bundled pre-optimized kernels.

Demo Booth Presentations:

Exhibit Hours:
Monday, September 26, 2016 09:30 - 16:30
Tuesday, September 27, 2016 09:30 - 17:10
Wednesday, September 28, 2016 09:30 - 17:10

1. Hexacloud: Real-time 3D Surface Reconstruction and Space Digitization for Robotics and Domotics focused on Scene Understanding using multiple RGBD Cameras.
Presenter: Dr.David Gonzalez, Research Scientist, Intel Labs
Description: The introduction of 3D cameras in the market such as the Intel RealSense is changing the mobile devices, home automation and service robotics landscapes by enabling wide range of unseen technological possibilities. The most common geometric representations used in robotics and domotics (domestic automation) are categorically split into surface meshes and volumetric occupancy maps. Our work enables the generation of these representations with additional properties (considering the temporal dimension) and improved performance (computing resources and memory footprint) beyond of the state-of-the-art. This is illustrated in a real-time demonstration with tools for each representation category.

2. 10bit HW HEVC 4K Playback
Presenter: Durgaprasad Bilagi, Media Architect, Visual and Parallel Computing Group (VPG)
Description: The demo showcases 10-bit HEVC 4K support in the latest 7th generation Core™ Processors. It is fully hardware accelerated and brings 4Kp60 experience in small form factors.

3. Computer Vision on Intel Embedded Platform
Presenter: Dukhwan Kim, Computer Vision Software Engineering Manager, ICG
Description: This demo showcases two computer vision algorithms (face detection, pedestrian detection) running through OpenCL and OpenVX on Intel Embedded Platform.

4. Intel Computer Vision SDK
Presenter: Jeffrey Mcallister, Technical consulting engineer, Software Solutions Group (SSG)
Description: This demo showcases Intel® Computer Vision SDK based on OpenVX*, which lets you harness the performance of computer vision accelerators from Intel in your processing pipeline.

Deep Learning for Image Processing & Computer Vision with MATLAB

Deep Learning for Image Processing & Computer Vision

Deep learning applications has rapidly evolved over the past decade and is now being used in fields varying from autonomous systems to medical image processing. This tutorial will cover both Machine learning and Deep learning techniques to help solve problems such as object detection, object recognition and classification.

This session will cover the following demonstrations:

1. Machine Learning Techniques for Scene Classification: We will explore new feature extraction methods available in our products, such as Bag of Visual Words. We will explore creating a classifier and which techniques work best for scene classification, such as K-NN and SVM classification techniques.

2. Creating a CNN from scratch: In this example, we will explore how to access and manage large sets of images, even when they do not fit into memory. We will create a 10-category classifier from scratch, defining the layers of a network and discuss parameter training options.

3. Transfer Learning: An alternative approach to training a model from scratch can be using a pre-trained model and re-training to perform a new classification task. We will show how to perform transfer learning in MATLAB and run the classifier using a live video feed.

4. Using a pre-trained CNN as a feature extractor: We will show leveraging the use of pre-trained networks for feature extraction. The pre-trained network can be a robust and powerful feature extractor, and we will compare this accuracy to a traditional feature extraction method as see in the first demo.

*All of the code used in the examples will be freely available to all attendees