Download the Technical Program ! NEW: UPDATED ON 11th FEBRUARY 2018 !
MCV 2018 includes three oral sessions and one session for demos and posters.
(1) Mathematics for Machine Learning.
In his seminal paper Computing Machinery and Intelligence (Mind, 1950), Alan Turing asks “Can machines do what we (as thinking entities) can do?”. Machine Learning is a field of artificial intelligence, addressing Turing’s question by a set of human-inspired techniques, today widely applied in many Computer Science applications. The first talk of this Section will present the mathematical fundations of Machine Learning for Computer Vision, while the second talks will discuss Deep Learning methods for the people/object tracking purpose in complex scenes with particular attention to mathematical insights.
Prof. Simone Bianco from Università degli Studi di Milano Bicocca, Italy
Dr. Oswald Lanz, from Fondazione Bruno Kessler, Trento, Italy
(2) Mathematics & Cinema There is a famous aphorism by Frank Capra: “Cinema is one of the three universal languages. The others two are mathematics and music”. What’s the relationship between Mathematics and Cinema? How does mathematics help cinema? The first talk will give an overview about mathematics in cinema, while the second talk will introduce a mathematical model for the restoration of degraded video sequences presenting regions where the original data are entirely lost.
Prof. Michele Emmer from Università La Sapienza di Roma, Italy
Dr. Riccardo March from Istituto per le Applicazioni del Calcolo “Mauro Picone” (CNR), Roma, Italy
(3) Mathematics for Human-inspired Color Image Processing.
From L'Allegria - da Ultime”, Milano 1914/1915, by Giuseppe Ungaretti: “Tappeto. Ogni colore si espande e si adagia/negli altri colori/ Per essere più solo se lo guardi” (“Carpet. Every color expands and lays down/ on the other colors/ to become more alone if you look at it”). How do humans see colors” As suggested by Ungaretti’s poetry, human color vision is a local spatial process. The first talk of this Section will discuss the need of processing jointly spatial and color information, both in human and machine vision, while the second talks will present some mathematical, human-inspired models for spatial color processing approaches devoted to image enhancement.
Prof. Alessandro Rizzi from Università degli Studi di Milano, Italy
Ms. Michela Lecca from Fondazione Bruno Kessler, Trento, Italy
Demo Session will include two invited demos, showing practical applications of machine learning and 3D techniques:
(1) Recognition and description of visual data using Convolutional Neural Networks (CNNs)
CNNs are widely used for image recognition and tagging. In this demo we will show some applications of CNNs that we have designed to complement our image tagging system. Specifically, we will show how it is possible to estimate both image quality and aesthetics for generic images, and how our image tagging system can be customized to particular domains such as: food recognition and retrieval, painter recognition and categorization. Finally, as a compendium of a face detection and recognition system, we will show the effectiveness of our CNN-based software for the simultaneous estimation of multiple facial attributes. The demos will be running on standard benchmark datasets and, when possible, on generic images chosen by the users.
Keywords: Convolutional Neural Networks, Blind image quality assessment, Image aesthetic assessment, Facial attributes
Demonstrator: Prof. Gianluigi Ciocca, Università di Milano Bicocca, Milano, IT
(2) 3D Reconstruction in your pocket
This demo will show how we can create digital replicas of real-world objects using commercial smartphones. With the smartphone camera pointed towards an object to digitally replicate, an app automatically selects and sends a subset of captured images to a remote 3D reconstruction server. The app shows an incrementing digital replica of the current reconstruction in real time to the user. 3D reconstruction can run concurrently and collaboratively with other users to speed up the acquisition of large objects, e.g. buildings.
Keywords: Image filtering, Structure from Motion, Epipolar Geometry, Optimisation.
Demonstrators: Dr. Fabio Poiesi, Fondazione Bruno Kessler, Trento, IT
Participants are encouraged to present a research in poster format. Posters may address issues related to the topics addressed in the oral sections, but they are not limited to.