Marco Pedersoli
Assistant Professor
Department of Electrical Engineering, ETS Montreal
Phone: +1 (514) 396-8742
Office: A-3480
marco.pedersoli@etsmtl.ca
Since February 2017 I am assistant professor at École de technologie supérieure (ETS), the youngest and fastest-growing university located in the center of Montreal, the AI Mountain!
+ Teaching
- ETS: “Réseaux de Neurones et Intelligence Artificielle” Autumn 2018
- ETS: “Apprentissage Machine” Hiver 2018
- ETS: “Réseaux de Neurones et Intelligence Artificielle” Autumn 2017
- KU Leuven: Embedded Systems and Multimedia 2014
- Teachning assiatant in UAB: Computer Science 2009-2012
- Teachning assiatant in UAB: Bioinformatics 2009-2012
- Teachning assiatant in UAB: Computational Logic 2008-2009
+ Original articles in refereed journals and books chapters
«Incremental multi-target domain adaptation for object detection with efficient domain transfer»Le Thanh Nguyen-Meidine, Madhu Kiran, Marco Pedersoli, Jose Dolz, Louis-Antoine Blais-Morin, Eric Granger” |
«Deep clustering: On the link between discriminative models and K-means»Mohammed Jabi, Marco Pedersoli, Amar Mitiche, Ismail Ben Ayed” |
«Self-paced and self-consistent co-training for semi-supervised image segmentation»Ping Wang, Jizong Peng, Marco Pedersoli, Yuanfeng Zhou, Caiming Zhang, Christian Desrosiers” |
«Automation of video-based location tracking tool for dairy cows in their housing stalls using deep learning»A. Zambelis, M. Saadati, G. M. Dallago, P. Stecko, V. Boyer, J. P. Parent, M. Pedersoli, E. Vasseur” |
«Deep co-training for semi-supervised image segmentation»Jizong Peng, Guillermo Estrada, Marco Pedersoli, Christian Desrosiers” |
Read More >
+ Papers in refereed conference proceedings
«F-CAM: Full resolution class activation maps via guided parametric upscaling»Soufiane Belharbi, Aydin Sarraf, Marco Pedersoli, Ismail Ben Ayed, Luke McCaffrey, Eric Granger” |
«A joint cross-attention model for audio-visual fusion in dimensional emotion recognition»R. G. Praveen, W. C. de Melo, N. Ullah, H. Aslam, O. Zeeshan, T. Denorme, M. Pedersoli, A. L. Koerich, S. Bacon, P. Cardinal, E. Granger” |
«Temporal stochastic softmax for 3D CNNs : An application in facial expression recognition»Theo Ayral, Marco Pedersoli, Simon Bacon, Eric Granger” |
«On the texture bias for few-shot CNN segmentation»Reza Azad, Abdur R. Fayjie, Claude Kauffmann, Ismail Ben Ayed, Marco Pedersoli, Jose Dolz” |
«Learning data augmentation with online bilevel optimization for image classification»Saypraseuth Mounsaveng, Issam Laradji, Ismail Ben Ayed, David Vazquez, Marco Pedersoli” |
The current bottleneck in deep learning is not much about the amount of available data, but rather the capability to process this data and the cost of annotating it.
My main objectives are:
- investigate and develop methodologies to reduce the computational cost of modern visual recognition techniques and models.
- find strategies and new algorithms to improve these methods performance on limited training data and/or annotations.
This will open the doors to the deployment of modern computer vision algorithms on the increasingly demanding market of small and computation limited portable and embedded devices.
+ Research Projects
Deep Learning with reduced Supervision: Deep learning requires Big data, but what about annotations? Do we really need each sample to be annotated? How far can we go with a reduced set of annotations? Can we compensate the lack of annotations with more computation? Is it better to use a few clean annotations or more but noisy annotations?
+ Papers
Learning by Exploration: Most of the common datasets used in Computer Vision are composed of samples (e.g. images) and labels (e.g. image categories). This is an ideal case that makes training and evaluation clear and simple. However, in the real world often data comes from an environment that should be explored. Here then new, yet very interesting problems appear. How to select from which data to learn? How and when to use supervision if the size of the environment to explore is exponentially large? How to evaluate the performance?
+ Papers
Efficient Computation: Datasets are getting larger and predictive models are becoming computationally very expensive. In this research line we study and propose methods for reducing the computational cost of deep learning models at both training and inference time.
+ Selected Publications
- Adversarial Learning of General Transformations for Data Augmentation, Saypraseuth Mounsaveng, David Vazquez, Ismail Ben Ayed, Marco Pedersoli, in ICLR Workshop, 2019. (pdf)
- An Attention Model for group-level emotion recognition, Aarush Gupta, Dakshit Agrawal, Hardik Chauhan, Jose Dolz, Marco Pedersoli, in ACM International Conference on Multimodal Interaction, October 2018. (pdf)
- Areas of Attention for Image Captioning, Marco Pedersoli, Thomas Lucas, Cordelia Schmid, Jakob Verbeek, in International Conference of Computer Vision, October 2017. (pdf)
- DeepProposal: Hunting Objects and Actions by Cascading Deep Convolutional Layers, Amir Ghodrati, Ali Diba, Marco Pedersoli, Tinne Tuytelaars, Luc Van Gool, in International Journal of Computer Vision, March 2017. (pdf,iccv15,code)
- Towards Automatic Image Editing: Learning to See another You, Xu Jia, Amir Ghodrati, Marco Pedersoli, Tinne Tuytelaars in BMVC, September 2016. (pdf,extended abstract,code)
- Learning where to position parts in 3D, Marco Pedersoli, Tinne Tuytelaars, in ICCV, December 2015. (pdf,3DV14,poster,code)
- Swap Retrieval: Retrieving Images of Cats When the Query Shows a Dog, Amir Ghodrati, Xu Jia, Marco Pedersoli, Tinne Tuytelaars, in ICMR, June 2015. (pdf)
- Combining where and what in change detection for unsupervised foreground learning in surveillance, Ivan Huerta, Marco Pedersoli, Jordi Gonzàlez, Alberto Sanfeliu, in Pattern Recognition,Vol.48(3), 709-719, 2015 (pdf)
- Weakly Supervised Object Detection with Convex Clustering, Hakan Bilen, Marco Pedersoli, Tinne Tuytelaars, in CVPR, June 2015. (pdf)
- A Coarse-to-fine approach for fast deformable object detection, Marco Pedersoli, Andrea Vedaldi, Jordi Gonzàlez, in Patter Recognition, Vol.48(5), May 2015. (pdf,cvpr11,pptx,code)
- Face detection without bells and whistles, Markus Mathias, Rodrigo Benenson, Marco Pedersoli, Luc Van Gool, in ECCV, September 2014. (pdf,odp,project)
- Is 2D information enough for viewpoint estimation?, Amir Ghodrati, Marco Pedersoli, Tinne Tuytelaars, in BMVC, September 2014. (pdf)
- Weakly Supervised Object Detection with Posterior Regularization, Hakan Binen, Marco Pedersoli, Tinne Tuytelaars, in BMVC, September 2014. (pdf)
- Using a deformation field model for localizing faces and facial points under weak supervision, Marco Pedersoli, Tinne Tuytelaars, Luc Van Gool, in CVPR, June 2014. (pdf,video1,video2,video3,code)
- Object Classification with Adaptable Regions, Hakan Bilen, Marco Pedersoli, Vinay P. Namboodiri, Tinne Tuytelaars, Luc Van Gool, in CVPR, June 2014. (pdf)
- An Elastic Deformation Field Model for Object Detection and Tracking, Marco Pedersoli, Radu Timofte, Tinne Tuytelaars, Luc Van Gool, International Journal of Computer Vision, June 2014, (pdf,report)
- Toward Real-Time Pedestrian Detection Based on a Deformable Template Model, Marco Pedersoli, Jordi Gonzàlez, Xu Hu, Xavier Roca, IEEE Transactions on Intelligent Transportation Systems, 15(1), September 2013. (pdf)
- Hierarchical Multiresolution Models for fast Object Detection, Marco Pedersoli, Phd Thesis, September 2012. (pdf,bib)
- For the complete list of my publications check my google scholar.
In 2015-2016 I have been post-doc in THOTH at INRIA Grenoblewith Dr. Cordelia Schmid and Dr. Jakob Verbeek. From 2012 to 2015 I was in VISICS at KU Leuven with prof. Tinne Tuytelaars. I obtained my Ph.D. at the Computer Vision Center and the Autonomous University of Barcelona (UAB) under the supervision of Jordi Gonzàlez and Juan José Villanueva. For more details check my CV.
+ Code
- Roi-Pooling in Lasagne. Porting of R-CNN roi-pooling in Lasagnegithub.
- Weakly Supervised Detection. Python (Caffe) Code based on Fast RCNN for weakly supervised detection github.
- Face detection evaluation code. Python and C source codebitbucket.
- Fast 3D Object Detection. Python and C source code github.
- Deformation Field Model. Python and C source code github.
+ Research Associate
- PhD
- Saypraseuth Mounsaveng
- Akhil P M
- Jizong Peng
- MsC
- Kristof Boucher Charbonneau
- Théo Arial
- Masih Aminbeidokhti
- Mirmohammad Saadati
+ Research Associate
- Mehraveh Javan
Marco
Pedersoli
Research & Innovation
Contact Us
Pavillon Principal (A)
1100, rue Notre-Dame Ouest
Montréal, QC, H3C 1K3
Room A-3600
Tel.: +1 (514) 396-8650
E-Mail: eric.granger@etsmtl.ca