LAMP

Our paper ‘On-the-fly Network Pruning for Object Detection‘ has been accepted at ICLR 2016 on the Workshop Track.

Also, the paper ‘Hierarchical Part Detection with Deep Neural Networks‘ has been accepted at the ICIP 2016.

The work ‘Fitnets: Hints for thin Deep Nets’ will be presented at the ICLR 2015. The paper proposes a framework to compress wide and deep networks into thin and deep ones.

Our paper ‘Accurate stereo matching by two-step energy minimization’ has been accepted in IEEE TIP (pdf). Sofware can be downloaded here and comparison on the Middlebury data set is vailable here.

Our work on discriminative sparse reconstruction for ranking problems like person re-identification has accepted in PAMI. This paper shows how to extend sparse discriminative classifiers to ranking problems through iterative re-weighting of sparse solutions. Code is available here.

The Learning and Machine Perception (LAMP) team at the Computer Vision Center conducts fundamental research and technology transfer in the field of machine learning for semantic understanding of visual data. The group works with a wide variety of visual data sources: from multispectral, medical imagery and consumer camera images, to live webcam streams and video data. The returning objective is the design of efficient and accurate algorithms for the automatic extraction of semantic information from visual media.

Our work on unsupervised learning of sparse features has been accepted in PAMI. The method aims at both lifetime and population sparsity in order to learn discriminative features. The code is available here and additional results are reported in this paper presented at WHISPERS.

Our BMVC paper on facial behavior analysis has been accepted for oral presentation (project page).

• Two new journals accepted: “Semantic Pyramids for Gender and Action Recognition” was accepted in IEEE Transactions on Image Processing, and“Leveraging local neighborhood topology for large scale person re-identification” will appear in Pattern Recognition.
• Joost van de Weijer is invited professor at Laboratoire Hubert Curien at the University of Jean Monnet St. Etienne for the month of June.

The LAMP Research Team has been officially recognized as an Emerging Research Group by the Generalitat de Catalunya! This is the first step in the process of becoming a Catalan Consolidated Research Group.

We have three papers accepted for presentation at ICPR 2014 (one oral and two posters). In “Scale Coding Bag-of-Words for Action Recognition” we propose two techniques for relative scale coding in the BOW pipeline; in“Unsupervised scene adaptation for faster multi-scale pedestrian detection” we show how scene-dependent pedestrian detector optimizations can be automatically mined from running detectors; and in “Fisher vectors over random density forests for object recognition” we propose construction of Fisher vectors for image recognition from random forst density estimators.