Raphaël Marée

Projects & Research Interests

2005-today: Bioinformatics and modeling, bioimage informatics

Since 2005, I am the GIGA Bioinformatics platform manager (scientific head: Prof. L. Wehenkel). In close collaboration with the Bioinformatics and Modeling research unit, we offer software development and data analysis services to academic (within the GIGA research center and beyond) as well as to industrial researchers. These services include classification of biological/biomedical data (SELDI mass spectra, microarrays, clinical databases, microscopy images, ...) obtained from various medical instrumentation based on machine learning methods.

In this context, my personal research interests lie within the field of "bioimage informatics", ie. the development and application of machine learning and computer vision methods to recognize cells, tissues and other biological or biomedical "objects" in large-scale, imaging datasets, and on the delivery of user-friendly softwares to help biologists/pathologists/etc. to facilitate the exploration of their imaging data, derive new knowledge, and support their findings and diagnoses.

In that respect, since October 2010 till 2013, I am the scientific coordinator of the CYTOMINE research project funded by the Walloon region (DGO6) which main goal is to develop a rich internet application for user-friendly and remote visualization, collaborative annotation, and automated analysis and quantification of high-resolution/high-throughput bioimages in cancer research and diagnostics (also known as Virtual Microscopy, Whole-slide imaging, Digital Pathology), in collaboration with researchers from GIGA-Cancer research unit and from the Department of Pathology at Erasme University Hospital.

I'm also actively involved in zebrafish image analyses for (see e.g. our paper at PRIB 2011) in collaboration with researchers from the GIGA-Development research unit.

                   

2002-today: Machine learning and computer vision

This research is about the design of generic methods for automatic image classification, retrieval, and semantic segmentation. Indeed, as potential applications of image recognition technologies are multidinous, we seek to develop general-purpose methods for the recognition of various types of images that share some visual regularities, without relying on too strong assumptions about patterns to recognize and acquisition conditions, and without having to rely on domain experts to design specific features.

In 2003, we proposed to combine ensemble of randomized decision trees with random extraction of subwindows (square patches) described by their raw pixel values for image classification/categorization. It was first described in our JDS03 paper (in french) and then in our SGAI-AI-2003 paper. The main contributions are summarized in the CVPR05 paper. See our other publications for details, various applications (in particular on biomedical problems), and latest extensions e.g. for distributed and incremental content-based image retrieval (MIR10) and semantic segmentation (VISAPP09, CVPR09-OTBVS). The java software PiXiT implements the CVPR05 method for image classification and I suggest to use it with its default parameters as a baseline method on new datasets. If results are satisfactory, then you do not need to develop a new method. ;-)

Currently (2011-...), we are conducting large-scale empirical studies on many datasets to better identify influential design choices and draw general guidelines for future use, and to further increase robustness of the approach. On several datasets, results are significantly improved compared to our previous works (paper in preparation).

Simple examples of image classification benchmarks I have been working on with the same approach:

>

Publications

To browse other publications from our group, please go here.

To access my publications on the institutional repository (ORBI), please go here.

Software

PiXiT, a Java software which implements the CVPR05 aforementioned image classification method is available upon request for evaluation and non-commercial purpose, in collaboration with PEPITe. Newer extensions and improvments are not yet included in the free evaluation version.

Conference Calendar

For years, for personal usage I'm trying to maintain a unofficial, browsable, conference calendar in the fields of machine learning, computer vision, biomedical imaging (or import the .ics iCalendar file in your application).
Tip: click on too see the calendar by month.

In the field of computer vision, the conference listing from USC is far more complete.

Étudiants

Propositions de sujets de TFE pour l'année académique 2011-2012 (new)

Misc

I listen very much to music (mostly ambient/electronica/fields recordings). I used to play selections on a local radio and co-organized the Panoptica festival with friends. I also enjoy reading (society essays and "independent" comics), eating (from thaï/indian/lebanese/japanese food to boulet-frites), traveling, taking pictures, ...

(Korea) (San Diego) (Beijing) (Tokyo-Kyoto) (Lisbon) (Brittany)