machine learning

I got my BSc in Computer Science in 2004 from the Universitat Autònoma de Barcelona (UAB), and my MSc and PhD in 2006 and 2010, respectively, both from the Computer Vision Center (Advanced Driver Assistance Systems group) in the UAB. My PhD was on Pedestrian Detection for driver assistance, a field that a few years later would become widely known autonomous driving.

I made research stays in the University of Surrey in Guildford (UK) and the Laboratoire d’Analyse et d’Architecture des Systèmes (LAAS-CNRS) in Toulouse (France), and a post-doc in the Computer Vision and Active Perception Lab (CVAP) of the KTH Royal Institute of Technology in Stockholm (Sweden). In these stages I researched on different topics such as classification, tracking and video understanding.

Then I moved back to Barcelona to join the start-up Catchoom at the end of 2014. We aimed at using deep learning classifiers to provide rich attribution to fashion images. In 2020 we fused with Slyce from USA and Humai from Austria to create Partium, where I currently work as a R&D Lead. We aim at retrieving industrial parts by using image and text to help operators in different industrial sectors at finding these parts.

Linkedin | Resume | CV



Industrial Spare Parts Retrieval (at Partium)

Partium consists in the retrieval of industrial sparse parts in several sectors such as engineering, railway, automotive, energy or home improvement aimed at improving maintenance and service. My work in this product involve leading several ML teams where we use Deep Learning of different modalities (text, image, etc.),

Garment Attribution (at Catchoom)

Deep Products consisted in the rich attribution of fashion imagery for e-shops. It involved the development of a Deep Learning framework from scratch, all the data management process, recruiting a team and researching ways to improve the performance of the attribution using different DL techniques.

Forensic Video Mining (at KTH)

Researching computer vision and machine learning techniques to perform image retrieval in surveillance imagery based on visual examples or textual-queries in the context of forensic investigations.

References: ICPR 2014.

Education Project – Traffic Sign Detection (at UAB)

Graduate course on computer vision aimed at detecting and recognizing traffic signs in videos. The ideas, program and materials of the paper were successfully in the 4 months “Project in CV” subject of the term 2011-2012 edition of the Master in CV and AI (UAB).

References: IEEE Trans. Education 2013.

Pedestrian Tracking (at LAAS)

Multi-person tracking can be exploited in applications such as driver assistance, surveillance, multimedia and human-robot interaction. With the help of human detectors, particle filters offer a robust method able to filter noisy detections and provide temporal coherence.

Results: PETS and TUD videos.
References: ACIVS 2012.

Pedestrian Protection for Driver Assistance (at CVC)

A PPS detects the presence of both static and moving pedestrians in front of the vehicle in order to warn the driver (acoustically, visually, etc.) and perform active actions on the host vehicle (automatic braking, evasive actions, etc.).

Results: Seq 6, Seq 9.
References: PAMI 2010, CVIU 2010, IbPria 2007, CVPR 2010, PAMI 2014.
Theses: MSc Thesis 2006, PhD Thesis 2010 (Slides).

Camera-road Pose Estimation (at UAB)

This was the topic of my BSc Thesis and continued with this work together with some fellows during my PhD. The knowledge of the road position is crucial for any vision-based driver assistance.

Results: Plane estimation.
References: ITS 2008, EL 2006.
Theses: BSc Thesis 2004.

Teaching and Advising

I taught both BSc and MSc courses, advised several MSc Theses and also professional internships both in the university and in the private company. Please check my CV for more info.

During 2019 and 2020 I taught these MSc lectures, including theory and a couple of pytorch assignments: Lectures on Machine Learning (MSc in IoT for e-Health – Smart Data Knowledge and Analytics). Here you can find the slides I prepared: