(Past) Vision Student Talks [ViST]

Past presentations :
(click on 'details' to show/hide abstract, slides, etc.)

December 3rd : Rocio Cabrera-Lozoya (Asclepios) : Image-based modelling: an application to cardiac ventricular tachycardia inducibility

Computational biomedicine is concerned with the development of integrative computer models of human biology at all levels of biological organization. There are three methods in which imaging data are used to construct a personalized model of an individual's physiology: 1) modelling the anatomy of a specific domain (organ) and specific subdomains (tissue types) using data obtained from static imaging; 2) defining the boundary and initial conditions of a model using dynamic imaging; 3) using imaging to characterize individual structural and functional properties of a tissue. In the context of ventricular tachycardia inducibility, using image-based modelling and a clinical approach, the spatial heterogeneity of the electrical properties of the left ventricular endocardium are studied to help predict the location of crucial entry/exit points of re-entrant ventricular tachycardia circuits.
November 26th : Serhan Cosar (Stars) : Data Clustering: A very brief overview

Clustering is a statistical data analysis technique that can be used in many fields including image processing, machine learning, bioinformatics. In this talk, I will try to give an overview of data clustering approaches. In particular I will try to answer the following questions: what is data clustering, when do we need it, and how do we do it?

[ slides ]
November 19th : Herve Lombaert (Asclepios) : Functional Maps: A Flexible Representation of Maps Between Shapes, Ovsjanikov et al, ACM Trans. on Graphics, 2012

How to represent whole correspondence maps with a simple compact matrix? Perhaps shapes could be even matched without finding a direct point-to-point correspondence - This paper proposes to generalize the notion of correspondence to a simpler, compact representation that maps functions on one shape onto another. Beyond matching shapes, this generalization allows straightforward transfer of functions (e.g., simple transfers of segmentations between shapes). Under the hood is the fascinating Laplacian decomposition of shapes.

Paper : [ original ] [ local copy ]

[ slides ]
November 5th : Marc-Michel Rohe (Asclepios) : Presentation of the paper Robust Principal Component Analysis

Abstract :
This article is about a curious phenomenon. Suppose we have a data matrix, which is the superposition of a low-rank component and a sparse component. Can we recover each component individually? We prove that under some suitable assumptions, it is possible to recover both the low-rank and the sparse components exactly by solving a very convenient convex program called Principal Component Pursuit; among all feasible decompositions, simply minimize a weighted combination of the nuclear norm and of the ℓ1 norm. This suggests the possibility of a principled approach to robust principal component analysis since our methodology and results assert that one can recover the principal components of a data matrix even though a positive fraction of its entries are arbitrarily corrupted. This extends to the situation where a fraction of the entries are missing as well. We discuss an algorithm for solving this optimization problem, and present applications in the area of video surveillance, where our methodology allows for the detection of objects in a cluttered background, and in the area of face recognition, where it offers a principled way of removing shadows and specularities in images of faces.

[ source ] [ local copy ]
October 29th : Carolina Garate Oporto (Stars) : Video-Based Human Behavior Understanding : A Survey

Abstract of the paper
Understanding human behaviors is a challenging problem in computer vision that has recently seen important advances. Human behavior understanding combines image and signal processing, feature extraction, machine learning, and 3-D geometry. Application scenarios range from surveillance to indexing and retrieval, from patient care to industrial safety and sports analysis. Given the broad set of techniques used in video-based behavior understanding and the fast progress in this area, in this paper we organize and survey the corresponding literature, define unambiguous key terms, and discuss links among fundamental building blocks ranging from human detec- tion to action and interaction recognition. The advantages and the drawbacks of the methods are critically discussed, providing a comprehensive coverage of key aspects of video-based human behavior understanding, available datasets for experimentation and comparisons, and important open research issues.

[ article (pdf) ]
October 22nd : Thomas Demarcy (Asclepios) : Pico Lantern and Megalodon

Presentation abstract :
In this talk, I will first present a quick overview of medical augmented reality. Illustrated by the following paper I will then discuss the need for depth perception in operating room and introduce surface reconstruction techniques. In the end of the talk I will introduce pico's adversary : the megalodon.

Paper abstract :
The Pico Lantern is proposed as a new tool for guidance in laparoscopic surgery. Its miniaturized design allows it to be picked up by a laparoscopic tool during surgery and tracked directly by the endoscope. By using laser projection, different patterns and annotations can be projected onto the tissue surface. The first explored application is surface reconstruction. The absolute error for surface reconstruction using stereo endoscopy and untracked Pico Lantern for a plane, cylinder and ex vivo kidney is 2.0 mm, 3.0 mm and 5.6 mm respectively. The absolute error using a mono endoscope and a tracked Pico Lantern for the same plane, cylinder and kidney is 0.8mm, 0.3mm and 1.5mm respectively. The results show the benefit of the wider baseline produced by tracking the Pico Lantern. Pulsatile motion of a human carotid artery is also detected in vivo. Future work will be done on the integration into standard and robot-assisted laparoscopic surgery.

[ article (pdf) ] [ slides (pdf) ]
October 15th : Seong-Gyun Jeong (Ayin) : Marked Point Process Model for Curvilinear Structures Extraction

In this paper, we propose a new marked point process (MPP) model and the associated optimization technique to extract curvilinear structures. Given an image, we compute the intensity variance and rotated gradient magnitude along the line segment. We constrain high level shape priors of the line segments to obtain smoothly connected line configuration. The optimization technique consists of two steps to reduce the significance of the parameter selection in our MPP model. We employ Monte Carlo sampler with delayed rejection to collect line hypotheses over different parameter spaces. Then, we maximize the consensus among line detection results to reconstruct the most plausible curvilinear structures without parameter estimation process. Experimental results show that the algorithm effectively localizes curvilinear structures on a wide range of datasets.

[ article (pdf) ] [ slides (pdf) ] [ slides (pptx) ]
October 8th : Bishesh Khanal (Asclepios) : Presentation of the paper Simulating Neurodegeneration through Longitudinal Population Analysis of Structural and Diffusion Weighted MRI Data

Abstract of the paper
Neuroimaging biomarkers play a prominent role for disease diagnosis or tracking neurodegenerative processes. Multiple methods have been proposed by the community to extract robust disease specific markers from various imaging modalities. Evaluating the accuracy and robustness of developed methods is difficult due to the lack of a biologically realistic ground truth. We propose a proof-of-concept method for a patient- and disease specific brain neurodegeneration simulator. The proposed scheme, based on longitudinal multi-modal data, has been applied to a population of normal controls and patients diagnosed with Alzheimer’s disease or frontotemporal dementia.We simulated follow-up images from baseline scans and compared them to real repeat images. Additionally, simulated maps of volume change are generated, which can be compared to maps estimated from real longitudinal data. The results indicate that the proposed simulator reproduces realistic patient-specific patterns of longitudinal brain change for the given populations.

[slides]
October 1st : Carlos Crispim-Junior (Stars) : Markov Logic and the Reasoning about Object Affordances in a Knowledge Base Representation

Presentation abstract:
I will be presenting the paper Reasoning about Object Affordances in a Knowledge Base Representation from Yuke Zhu, Alireza Fathi, and Li Fei-Fei (ECCV 2014).
Link: http://vision.stanford.edu/pdf/zhu14.pdf [or local copy here]
This paper presents a new representation for object affordance detection extracting together affordance label, human pose and relative position between object and person pose (e.g. next to) by the construction of a knowledge base and the usage of Markov logic formalism. I will give a brief introduction to Markov Logic formalism and its role in the approach proposed in this paper.

Paper abstract:
Reasoning about objects and their affordances is a fundamental problem for visual intelligence. Most of the previous work casts this problem as a classification task where separate classifiers are trained to label objects, recognize attributes, or assign affordances. In this work, we consider the problem of object affordance reasoning using a knowledge base representation. Diverse information of objects are first harvested from images and other meta-data sources. We then learn a knowledge base (KB) using a Markov Logic Network (MLN). Given the learned KB, we show that a diverse set of visual inference tasks can be done in this unified framework without training separate classifiers, including zero-shot affordance prediction and object recognition given human poses.

[slides]
September 24th : Effrosyni Doutsi (MediaCoding, I3S) : IEEE 1857 Standard : a next generation standard for capturing and coding surveillance videos which is recognition friendly for metadata analysis

The discussion of this week is about a recently released compression algorithm for surveillance videos named IEEE-1857. This algorithms has been built according to the specific constraints of video surveillance systems and it obtains high coding efficiency and low complexity comparing to the state-of-the-arts H.264/AVC and H.265/HEVC. This standard is the groundbreaking approach in the next generation smart video-surveillance systems which offers a powerful technology to support video analysis and recognition.

Paper : IEEE 1857 Standard Empowering Smart Video Surveillance Systems

[slides]
September 17th : Mehdi Hadj-Hamou (Asclepios) : Reducing the Dimensionality of Data with Neural Networks.

After describing what a neural network is, I will present a dimensionality reduction method based on neural networks (Hinton and Salakhutdinov, 2006).

Source: http://www.cs.toronto.edu/~hinton/science.pdf [Local copy here]
September 10th : Gaël Michelin (Morpheme) : Quantitative 4D analyses of epithelial folding during Drosophilia gastrulation

I will present a novel software called EDGE4D. This software enables quantification of the dynamics of cell shape changes during Drosophilia gastrulation. EDGE4D software could represent a major step for the understanding of dynamic tissue morphogenesis.

Source: https://sites.google.com/site/edge4dsupplement/

[slides]
September 3rd : Chloe Audigier (Asclepios) : Surrogates for image registration accuracy

This presentation will give a short introduction to image registration and will focus more on how to assess the accuracy of nonrigid image registrations. Surrogate measures commonly used such as tissue label, overlap scores, image similarity, image difference, or inverse consistency error will be presented. The talk will be illustrated with the CURT algorithm which shows great performances, but yet present some issues.

[ article ] [ slides (pptx) ]
July 16th : Nina Miolane (Asclepios) : Statistical Analysis on Manifolds : Generalizations of Principal Component Analysis

A wide variety of problems in Computer Vision possess a non linear structure and are therefore naturally modeled with Riemmanian geometry. Thus, statistical tools for manifolds data are needed. This presentation will give an overview of statistical analysis on manifolds. More precisely, we will introduce the generalizations of Principal Component Analysis and Probabilistic Principal Component Analysis to manifolds, which are: Principal Geodesic Analysis (linear and exact versions), Geodesic Principal Component Analysis and Probabilistic Principal Geodesic Analysis.
July 9th : Alexis Zubiolo (Morpheme) : Machine learning and artificial neural network for face verification

This presentation will give a motivation and a short introduction to machine learning and artificial neural nets techniques applied to face verification. The talk will be illustrated with the DeepFace algorithm which has shown great performances in this field.

Article : https://www.facebook.com/publications/546316888800776/

[slides]
July 2nd : Mathieu Le (Asclepios) : Monte Carlo

Based on the paper “Gaussian processes to speed up hybrid Monte Carlo for expensive Bayesian integrals”, I will present an overview of the Monte Carlo method, from the Metropolis-Hastings algorithm to the Hamiltonian Monte Carlo.

Papers :
Rasmussen, Carl Edward. Gaussian processes to speed up hybrid Monte Carlo for expensive Bayesian integrals. Bayesian Statistics 7: Proceedings of the 7th Valencia International Meeting. Oxford University Press, 2003.

Neal, Radford M (2011). MCMC Using Hamiltonian Dynamics. In Steve Brooks, Andrew Gelman, Galin L. Jones, and Xiao-Li Meng. Handbook of Markov Chain Monte Carlo. Chapman and Hall/CRC.

Slides : [pdf] or [pptx].
June 25th : Ihsen Hedhli (Ayin) : New cascade model for hierarchical joint classification of multitemporal, multiresolution and multisensor remote sensing data

In this talk, we propose a novel method for the joint classification of multidate, multiresolution and multisensor remote sensing imagery, which represents a vital and fairly unexplored classification problem. The proposed classifier is based on an explicit hierarchical graph-based model sufficiently flexible to deal with multisource coregistered time series of images collected at different spatial resolutions. An especially novel element of the proposed approach is the use of multiple quad-trees in cascade, each associated with each new available image at different dates, with the aim to characterize the temporal correlations associated with distinct images in the input time series. Experimental results are shown with multitemporal and multiresolution Pléiades data.
June 18th, Euler bleu : Antitza Dantcheva (Stars) : Soft Biometrics: Traits and Applications

This talk will explore the role of biometric data in providing information that deviates from the traditional realm of identification, and rather involves other personal and statistical characteristics, such as age, gender, ethnicity, gait, body height, facial hair, anthropometric measures and accessories. These characteristics, commonly referred to as soft biometrics, can be powerful tools in many applications. I will give an overview of such applications, related benefits - such as the fast and enrolment free analysis, and highlight state-of-the-art techniques.
June 11th : Loic Le Folgoc (Asclepios) : A true story of trees, forests & papers

From regression, pose estimation, image restoration to classification and clustering, object detection, density learning and coffee brewing: random forests are all over the place in machine learning, computer vision and medical imaging. In this talk I will give insight into how they work and why they are popular. Kinect is getting old, so let's see instead how Microsoft uses filter forests for turtle and elephant denoising in their brand new CVPR 2014 paper!

[slides] [article by Fanello et al.]
June 4th : Lola Xiomara Bautista Rozo (Morpheme) : Superresolution imaging for neuroscience

The purpose of this talk is to present a review of the basic principles and practical implementations of the new methods that had been developed to overcome the resolution barrier in microscopy, as well as a discussion of their potentials for neuroscience research and recent applications in neurobiology.

Reference: Jan Tønnesen, U. Valentin Nägerl. Superresolution imaging for neuroscience. Experimental Neurology 242 (2013) 33-40.

[slides] [article by Tønnesen & Nägerl]
May 28th : Vania Bogorny (Universidade Federal de Santa Catarina (UFSC), Brazil, invited by Stars) : An Overview of some Current Methods for Trajectory Data Analysis

In this talk I will give an overview of some of our current works being developed at UFSC on trajectory data analysis, mainly for pedestrian trajectories. First, I will present a recent Conceptual Data Model for Semantic Trajectories, where we add several types of features to trajectories. Second, I will present some algorithms for detecting specific behaviors such as meet, avoidance, and chasing.
May 21st : Jan Margeta (Asclepios) : Google’s house number recognition (with convolutional nets)

Deep learning with convolutional neural nets has recently seen renaissance in the field of machine learning from speech, image and video data. In this talk I will give a short intro on how convolutional neural nets work. We will see how well designed crowdsourcing can be used to get cheaply massive sets of training data and how neural nets are used for street number recognition in Street view at Google to achieve near human level performance.

[slides]
May 14th : Paula Craciun (Ayin) : Optical illusions and their influence on vision

Computer vision systems have been modeled after the human vision system since their emergence. The link between the two systems is obvious: digital cameras serve as the machine counterpart of the human eye, the algorithms and techniques used to extract meaningful information from data after image acquisition are machine substitutes of the human brain. Thus, a deeper understanding on how the human brain processes the visual information can lead to better, biology inspired algorithms in the domain of machine vision. One way to explore how the human brain interprets visual information is to identify when it fails and why. Optical illusions are typical cases where our brain gets tricked to see something that is not actually there. In this talk I will emphasize the link between machine vision and human vision and give a short overview on optical illusions as well as some reasons why the human vision system fails to recognize them.

[slides]
May 7th, Euler violet, 10:30 am : Marcelo Bertalmio (invited speaker: Universitat Pompeu Fabra, Barcelona, Spain) : From Image Processing to Computational Neuroscience: A Neural Model Based on Histogram Equalization

There are many ways in which the human visual system works to reduce the inherent redundancy of the visual information in natural scenes, coding it in an efficient way.
The non-linear response curves of photoreceptors and the spatial organization of the receptive fields of visual neurons both work towards this goal of efficient coding. A related, very important aspect is that of the existence of post-retinal mechanisms for contrast enhancement that compensate for the blurring produced in early stages of the visual process. And alongside mechanisms for coding and wiring efficiency, there is neural activity in the human visual cortex that correlates with the perceptual phenomenon of lightness induction.
In this talk I will present a neural model that is derived from an image processing technique for histogram equalization, and that is able to deal with all the aspects just mentioned: this new model predicts lightness induction phenomena, and improves the efficiency of the representation by flattening both the histogram and the power spectrum of the image signal.
April 30th : Minh Khue Tran Phan (Stars) : Serious Games and Interactive Systems

Serious Game gives today a new way for using Games. They make use of instructional and video game elements, most often, for nonentertainment purposes, like training, marketing, education purpose. We can find a lot of applications in many domains (for exemple : defense, healthcare, Industry, Marketing, Education, etc.). However, it still remains a game which requests many interactions between player and game. For some particular players (senior, patient), the use (control) of the game becomes difficult.
In this talk, first, i will give a brief history about Serious Games. Second, i will present my thesis work in the framework of a Serious Game for Alzheimer patient.

[slides and videos]
April 23rd : Nazre Batool (Ayin) : Random Field Models for Applications in Computer Vision

The goal of this talk is to review random field models for computer vision. Markov Random Field (MRF) models have been the most popular class of models for computer vision applications. Another class of models is based on Conditional Random Fields (CRF) which were introduced first for labeling 1-D sequences and then for 2-D images. I will present key differences between MRF and CRF highlighted on the basis of generative vs. discriminative probabilistic models.

[slides]
April 16th : Vikash Gupta (Asclepios) : MRI for fetal brain: A survey on motion correction and multi-modal registration with ultrasound images

Magnetic Resonance Imaging (MRI) is one of the most used non-invasive imaging modality. MRI is particularly useful in brain imaging for studying brain diseases. Diffusion weighted imaging (DWI) is an MR technique which can be used to track white matter fibers in the brain. However, these imaging modalities are often plagued by imaging artifacts introduced by the eddy currents, patient motion etc. The most common technique used for imaging the fetus is ultrasonography (US). However, MRI presents increased field of view (FOV), superior contrast of the soft tissues compared to US. Imaging the human fetus is particularly difficult because we do not have any direct control on the motion of fetus inside the mother. The motion artifacts hamper image interpretation and often requires a repeat scan in order to establish diagnosis. In the talk, we will discuss some of the motion correction algorithms used in context of fetal MRI. We will also discuss some techniques for multi-modal (MRI and US) registration techniques for fetal brain.

[article 1] [article 2] [slides]
April 9th : Ana Rita Lopes Simoes (Morpheme) : Detecting Pulse from Head Motions in Video, by the MIT Artificial Intelligence Lab, CVPR 2013

[local copy of the article] [slides] [video]
April 2nd : Pierre Kornprobst (NeuroMathComp) Partial Differential Equations (PDE) in image processing

[slides]
March 26th : Aurélie Boisbunon (Ayin) : Sparsity and image processing
The notion of sparsity has become quite "hype" in the Machine Learning and Statistics communities, and its usefulness is also clear in Image Processing.
But how does it translate to signals and images in practice?
I will review some basics on the subject of sparsity and key articles that stand at the boundary between Machine Learning and Image processing.

Some references:
- A Wavelet Tour of Signal Processing, by Stéphane Mallat
- Online dictionary learning for sparse coding, by J. Mairal, F. Bach, J. Ponce, G. Sapiro
- Enhancing Sparsity by Reweighted l1 Minimization, by Emmanuel J. Candès, Michael B. Wakin, Stephen P. Boyd
[slides]
March 19th : Michal Koperski (Stars) : Introduction to action recognition

I will discuss motivation for action recognition and possible applications. In my talk I will also cover selected approaches for action recognition like: Pose estimation for action recognition, Motion based methods and Space time methods. In my talk I would like to give you an idea of the problem statement, main challenges and proposed solutions, so you can relate them to the topic of your study.

[slides] [videos]
March 12th : Emmanuel Soubies (Morpheme) : Multiple Birth and Cut Algorithm for Multiple Object Detection

[partly based on this article] [slides]
March 5th : Sławomir Bąk : HOGgles : what the computer really sees through image descriptors

[original article: pdf] [slides]

Next presentations

Back to main page