Researcher at NUS and MIT
Researcher at NUS and MIT
I am a research fellow at NUS (faculty host Prof. Qi Zhao), and a research affiliate at MIT (faculty host Prof. Poggio). I obtained the doctorate from ETH Zurich in 2014. I was fortunate to receive the European Intel Doctoral Student Award for my thesis.
My colleagues and I develop computational models of human visual perception. This will help to understand the principles of human intelligence and learning, and design intelligent machines.
We work on the hypothesis that the computations done in the human and artificial visual systems, aim at reducing the amount of examples used to learn to recognize objects. We develop models based on our hypothesis, that will unify the different artificial visual systems, and also model the computations done in the human visual system. Namely, we are currently building and analyzing computational models of object recognition that take into account aspects of human visual attention and memory.
X. Huang, C. Shen, X. Boix, Q. Zhao. SALICON: Reducing the Semantic Gap in Saliency Prediction by Adapting Neural Networks ICCV 2015.
E. Ozkan, G. Roig, O. Goksel, X. Boix. Herding Generalizes Diverse M-Best Solutions
A. Volokitin, M. Gygli, X. Boix. Predicting When Saliency Maps are Accurate and Eye Fixations Consistent CVPR 2016.
L. Yan, X. Boix, G. Roig, T. Poggio, Q. Zhao. Foveation-based Mechanisms Alleviate Adversarial Examples
M. Jiang, X. Boix, J. Xu, G. Roig, L. Van Gool, Q. Zhao. Learning to Predict Sequences of Human Visual Fixations TNNLS 2016.
M. van den Bergh, X. Boix, G. Roig, L. Van Gool. SEEDS: Superpixels Extracted via Energy-Driven Sampling IJCV 2015.
OpenCV Challenge Award 2015
X. Boix*, J.M. Gonfaus*, J. van de Weijer, A. Bagdanov, J .Serrat, J. Gonzalez. Harmony Potentials IJCV 2012.
PASCAL VOC Segmentation Challenge Winner 2010.
G. Roig*, X. Boix*, R. De Nijs, S. Ramos, K. Kuhnlenz, L. Van Gool. Active MAP inference in CRFs for efficient semantic segmentation ICCV 2013.
X. Boix*, G. Roig*, S. Diether, L. Van Gool. Self-Adaptable Templates for Feature Coding NIPS 2014.
* means equal contribution
Current state-of-the-art techniques for object recognition seem to be very easy to fool by imperceptible perturbations of the image. In a recent paper we provide new insights, and we introduce a new research direction to solve this fundamental problem.
Record-breaking results on
predicting eye-fixation locations
Email: elexbb at nus dot edu dot sg
Address: NUS #E4-06-21, 4 Engineering Drive 3
Email: xboix at mit dot edu
Address: MIT 46-5189, 43 Vassar Street,
MA 02139, Cambridge, USA