Company Project Proposals


Note: Greyed Entries: already assigned projects.

Current year proposals

  2023-2024 proposals
  Real-Time Image Processing, and Information Extraction for AI-driven League of Legends Coaching
This project integrates computer vision with the analysis of League of Legends using the LoL-v2t dataset. It focuses on extracting real-time metrics, analyzing player behavior, and predicting in-game events. The project consists of three phases: Data Extraction, Real-Time Validation, and Predictive Modeling, spanning data transformation to implementing predictive algorithms. Check the extended abstract for more details.
Extended abstract: Download PDF
Academic Supervisor: Jordi Sanchez Riera
Supervisor e-mail: jsanchez@iri.upc.edu
Institution: UPC
Assigned Student Name: Rosana Valero Martínez
Student e-mail: Rosana.Valero@autonoma.cat
Company: Omashu
Contact e-mail: aitor@omashu.gg
Contact Person: Aitor Mier Pons
Confidential: No
Date: 2024-05-01 00:41:13
  Data Warehouse / Data governance and lifecycle management
Improfit is a company that develops solutions for fitness and rehabilitation based in computer vision for pose estimation where we monitor body movements. There are many data from the inference of all body points and the company wants to structure and create data strategy in order to work in its own algotythms and be able to train the new models with structured and structured data. The project is already in the market with some customers and the companies wants to achieve next steps.
Extended abstract: Download PDF
Academic Supervisor: Javier Laplaza
Supervisor e-mail: jlaplaza@iri.upc.edu
Institution: UPC
Company: improfit (wellfit tec sl)
Contact e-mail: andreu@improfit.ai
Contact Person: Andreu Casadella
Confidential: Yes
Date: 2024-04-17 17:34:43
  Pose estimation Computer vision project for people with low mobility : upper body and face
Project CV Low mobility : upper body and face Improfit is a company that develops solutions for fitness and rehabilitation based in computer vision for pose estimation where we monitor body movements. The project will be to check angles and evaluate movements for people with limited mobility. We will also develop face movements. At the beginning it will be single person and latter multiperson
Extended abstract: Download PDF
Academic Supervisor: Isamel Benito
Supervisor e-mail: ibenitoal@uoc.edu
Institution: UOC
Company: improfit (wellfit tec sl)
Contact e-mail: andreu@improfit.ai
Contact Person: Andreu Casadella
Confidential: Yes
Date: 2024-04-09 05:54:04
  MAMMOGRAPHY IMAGE ANALYSIS FOR BREAST CANCER RISK ASSESSMENT
The goal of the project is to develop fair artificial intelligence (AI) models for breast cancer screening and risk assessment scenarios. This project has a compensation of 7€ / hour
Extended abstract: Download PDF
Academic Supervisor: Karim Lekadir
Supervisor e-mail: karim.lekadir@ub.edu
Institution: UB
Company: Fundacion Vicomtech
Contact e-mail: magirreg@vicomtech.org
Contact Person: Mikel Agirre Garmendia/Karen Lopez-Linares
Confidential: No
Date: 2024-04-08 16:56:42
  Hands and body tracking through computer vision in real time
Project CV Hands and body tracking Improfit is a company that develops solutions for fitness and rehabilitation based in computer vision for pose estimation where we monitor body movements. The project will be to check angles and different positions of hands and other parts of the body and interact with objects integrated with augmented reality.
Extended abstract: Download PDF
Academic Supervisor: Simone Tassani
Supervisor e-mail: simone.tassani@upf.edu
Institution: UPF
Assigned Student Name: Marco Cordón Vaquero
Student e-mail: Marco.Cordon@autonoma.cat
pre-Assigned Student Name: marco cordon
Company: improfit (wellfit tec sl)
Contact e-mail: andreu@improfit.ai
Contact Person: Andreu Casadella
Confidential: No
Date: 2024-04-08 09:32:42
  3D Building Surface Reconstruction with Radiance Fields
This project focuses acquiring robust reconstructions of buildings in the context of virtual conference venues/omniconferences6 and historical buildings in indoor and outdoor scenes. Here we include the usage of constrained unconstrained structural priors as well as surface completion techniques to increase the robustness and feasibility of building reconstructions.
Extended abstract: Download PDF
Academic Supervisor: Antonio Agudo
Supervisor e-mail: antonio.agudo@upc.edu
Institution: UPF
Company: Eurecat - Centre Tecnologic de Catalunya
Contact e-mail: david.berga@eurecat.org
Contact Person: David Berga
Confidential: No
Date: 2024-03-27 15:07:28
  3D Model Editing with text-based transfer
In this project we will use text-based instructions for generating and editing NeRF and Gaussian Splatting representations. This focus will tackle the exploration of the capabilities on adapting and extending the current reconstruction algorithms and therefore developing and integrating end-to-end architectures that solve the possible problems encountered with distinct data characteristics.
Extended abstract: Download PDF
Academic Supervisor: Antonio Agudo
Supervisor e-mail: antonio.agudo@upc.edu
Institution: UPF
Company: Eurecat - Centre Tecnologic de Catalunya
Contact e-mail: david.berga@eurecat.org
Contact Person: David Berga
Confidential: No
Date: 2024-03-27 11:15:16
  3D Frame Tracer Using Multi-Camera System and Deep Learning for Eyeglass
Eyeglass lens tracing is a critical step in the optical industry to ensure that lenses fit precisely into the frame, providing comfort and optimal vision for wearers. Traditional frame tracers employ mechanical or 2D imaging techniques. However, recent advancements in computer vision and deep learning present an opportunity to enhance the accuracy and efficiency of frame tracing.
Academic Supervisor: Antonio Agudo
Supervisor e-mail: aagudo@iri.upc.edu
Institution: UPF
pre-Assigned Student Name: Manel Guzman
Company: Horizons Optical
Contact e-mail: contact@horizonsoptical.com
Contact Person: Manel Guzman
Confidential: Yes
Date: 2024-03-16 11:07:58
  New AI methods to ship detection from space
The objective of this work is to detect ships and other boats from space, implementing and studying different methodologies to lighten up the computation requirements while still meeting the quality factors. Embarking these algorithms in small satellites is a challenge due to the restrictive power capabilities of the computers in space.
Academic Supervisor: Felipe Lumbreras
Supervisor e-mail: felipe@cvc.uab.cat
Institution: UAB
co-Supervisor: Dani Ponsa
co-Supervisor e-mail: daniel@cvc.uab.cat
Company: Institut d'Estudis Espacials de Catalunya (IEEC)
Contact e-mail: marius.monton@ieec.cat
Contact Person: Marius Monton
Confidential: No
Date: 2024-03-08 17:35:36
  Data Augmentation for Earth Observation
Data augmentation techniques are used to artificially increase the size of a dataset by applying transformations to the original data. However, the transformations that are commonly used for natural images may not be suitable for EO data or yield the best results. In this project you will develop new augmentation tecniques for EO data and evaluate against state of the art data augmentation.
Extended abstract: Download PDF
Academic Supervisor: Robert Benavente Vidal
Supervisor e-mail: robert@cvc.uab.cat
Institution: UAB
Company: EarthPulse
Contact e-mail: juan@earthpulse.es
Contact Person: Juan B. Pedro
Confidential: No
Date: 2024-03-08 16:09:09
  Vision-Language Models for Earth Observation
Vision-language models are a type of model that can understand and generate natural language descriptions of visual content. These models have been shown to achieve state-of-the-art performance on a wide range of tasks, and have the potential to unlock the full potential of Earth Observation data by enabling new ways to interact with and understand the data.
Extended abstract: Download PDF
Academic Supervisor: Xim Cerda Company
Supervisor e-mail: xcerda@cvc.uab.cat
Institution: UAB
Company: EarthPulse
Contact e-mail: juan@earthpulse.es
Contact Person: Juan B. Pedro
Confidential: No
Date: 2024-03-08 16:04:56
  Data Fusion in Earth Observation
Data fusion is the process of combining data from multiple sources to produce more accurate, reliable, and complete information than can be obtained from any individual data source. In the context of Earth Observation, data fusion can be used to combine data from different sensors, modalities, and resolutions to produce more accurate and reliable information about the Earth's surface.
Academic Supervisor: Felipe Lumbreras
Supervisor e-mail: felipe@cvc.uab.cat
Institution: UAB
Company: EarthPulse
Contact e-mail: juan@earthpulse.es
Contact Person: Juan B. Pedro
Confidential: No
Date: 2024-03-08 16:03:44
  Foundation Models for Earth Observation
Foundation models are large neural networks that are trained in a self-supervised way on a diverse range of tasks and datasets, and are then fine-tuned on a specific task of interest. These models have been shown to achieve state-of-the-art performance on a wide range of tasks, and have the potential to unlock the full potential of Earth Observation data.
Extended abstract: Download PDF
Academic Supervisor: Joan Serrat
Supervisor e-mail: joans@cvc.uab.cat
Institution: UAB
Company: EarthPulse
Contact e-mail: juan@earthpulse.es
Contact Person: Juan B. Pedro
Confidential: No
Date: 2024-03-08 16:01:47
  Comparative Analysis of Artificial Intelligence techniques to asses a rare disease: Collagen VI-rela
Rare diseases are characterized by their low prevalence among the paediatric population, resulting in a limited number of patients and available medical samples. Besides, many rare diseases have a variable phenotype that difficult considerably their assessment, even for experienced clinicians, lasting at least five years to achieve an accurate diagnosis. Collagen VI-related Congenital Muscular Dys
Academic Supervisor: Josep Maria Porta
Supervisor e-mail: porta@iri.upc.edu
Institution: UPC
Assigned Student Name: Marcos Frías Nestares
Student e-mail: Marcos.Frias@autonoma.cat
pre-Assigned Student Name: Marcos Frias
Company: Hospital Sant Joan de Deu
Contact e-mail: monica.roldan@sjd.es
Contact Person: monica roldan
Confidential: Yes
Date: 2024-02-12 14:31:15
  Large vision-language models applied to video understanding and anomaly detection in the surveillanc
The project may involve, but not be limited to, large vision-language models applied to video understanding and anomaly detection in the surveillance context.. It will be part of the Trainee Program/Internship – Machine Learning Research that will align with the work you will do during your internship in Milestone.
Extended abstract: Download PDF
Academic Supervisor: Sergio Escalera
Supervisor e-mail: sescalera@ub.edu
Institution: UB
Company: Milestone Systems
Contact e-mail: receptionbarcelona@milestone.dk
Contact Person: Cecilia Bengtsson
Confidential: Yes
Date: 2024-01-09 10:25:47

  2022-2023 proposals
  Fish monitoring system with stereo camera on edge computing device
This thesis proposes the study and development of a fish monitoring system on an edge computing device, using a stereo camera integrated into a fishing net. Objectives include fish count, species classification, and size estimation. Deep learning models will be trained and adapted, leveraging the disparity information from the stereo images.
Extended abstract: Download PDF
Academic Supervisor: Antonio Agudo
Supervisor e-mail: antonio.agudo@upc.edu
Institution: UPF
Assigned Student Name: Alexander Tempelaar Sanchez
Student e-mail: Alexander.Tempelaar@autonoma.cat
pre-Assigned Student Name: Alexander Tempelaar Sanchez
Company: Coronis Computing, SL
Contact e-mail: contact@coronis.es
Contact Person: Rafael Garcia
Confidential: No
Date: 2023-05-31 11:43:44
  Lane Detection for Automated Driving
Development of algorithms for lane detection from images in the CULane dataset. In the first stage, the student will review the state of the art, and we will define the methodology to improve the state of the art during the second stage of the project.
Extended abstract: Download PDF
Academic Supervisor: Antonio Agudo
Supervisor e-mail: antonio.agudo@upc.edu
Institution: UPF
pre-Assigned Student Name: Sergi Vidal Bazan
Company: IDIADA Automotive Technology S.A.
Contact e-mail: monica.lores@idiada.com
Contact Person: sergi.vidal.bazan@idiada.com
Confidential: Yes
Date: 2023-05-22 16:47:22
  Prediction of Team Movement in Football
This project aims to explore different approaches for motion prediction of football player trajectories. We will investigate the use of MLP and transformer-based approaches to model the temporal and social interactions between agents in a football game.
Academic Supervisor: Francesc Moreno-Noguer
Supervisor e-mail: francesc.moreno.noguer@upc.edu
Institution: UPC
Assigned Student Name: Guillem Capellera Font
Student e-mail: Guillem.CapelleraFont@autonoma.cat
pre-Assigned Student Name: Guillem Capellera
Company: Kognia Sports Intelligence
Contact e-mail: antrubio10@gmail.com
Contact Person: Antonio Rubio
Confidential: Yes
Date: 2023-04-24 10:30:49
  Enhancing ADAS Object Detection through ISP Pipeline Optimization using Deep Learning
This master dissertation explores Deep Learning-based approaches to automate ISP tuning for automotive cameras, aiming to improve vision-based functionalities in AD/ADAS systems. The methodology involves using Deep Learning for RAW image processing to learn ISP parameters and benchmarking the performance of a Deep Learning-based vision algorithm.
Extended abstract: Download PDF
Academic Supervisor: Javier Vazquez
Supervisor e-mail: jvazquez@cvc.uab.cat
Institution: UAB
Assigned Student Name: Rachid Boukir Fatimi
Student e-mail: Rachid.Boukir@autonoma.cat
pre-Assigned Student Name: Rachid Boukir Fatimi
Company: FICOSA ADAS
Contact e-mail: miguel.rib@ficosa.com
Contact Person: Miguel Humberto Rib Perez
Confidential: Yes
Date: 2023-04-21 22:08:54
  Multiperson 3D pose estimation
Develop an algorithm for multi-person 3D pose estimation from single camera. Apply the algorithm to design an automatic trainer, able to count the repetitions of specific physical exercises and assess its quality on groups of 5 people approximately
Academic Supervisor: francesc moreno-noguer
Supervisor e-mail: francesc.moreno.noguer@upc.edu
Institution: UPC
pre-Assigned Student Name: Adam Szummer
Company: Wellfit Tec SL (improfit.ai)
Contact e-mail: andreu@improfit.ai
Contact Person: Andreu Casadella
Confidential: No
Date: 2023-04-19 13:03:27
  On the edge deep vision system for an efficient and precise detection of workpiece markings
The project will encompass the study and development of a detection system of on the edge markings focussed on computational efficiency and precision. This will require the study of large quantities of labelled data, the selection of optimum resolution strategies, implementation, trial and optimization of the chosen architectures, as well as the realisation of the results validation.
Extended abstract: Download PDF
Academic Supervisor: Ernest Valveny
Supervisor e-mail: ernest@cvc.uab.cat
Institution: UAB
Assigned Student Name: Iñigo Auzmendi Iriarte
Student e-mail: Inigo.Auzmendi@autonoma.cat
pre-Assigned Student Name: Inigo Auzmendi
Company: IDEKO S.COOP
Contact e-mail: notxoa@ideko.es
Contact Person: Nerea Otxoa Guridi
Confidential: Yes
Date: 2023-04-11 07:47:12
  Modern 3D material digitization for cultural heritage preservation
The field of 3D digital reconstruction has undergone a soaring and exciting boost with the recent invention of Neural Radiance Fields (NeRF). Within few years researchers have tightened the nuts to improve its photorealistic visual quality from multiple data types and optimized both training and inference time; and thus providing new solutions in several application fields.
Academic Supervisor: Joost Van de Eijer
Supervisor e-mail: joost@cvc.uab.es
Institution: UAB
Assigned Student Name: Advait Dixit
Student e-mail: Advait.Dixit@autonoma.cat
pre-Assigned Student Name: Advait Abhijeet Dixit
Company: Fundacio Eurecat
Contact e-mail: rafael.redondo@eurecat.org
Contact Person: Rafael Redondo
Confidential: No
Date: 2023-04-05 17:25:36
  Preserving the Past: Enhancing 3D geometry recovery for cultural heritage
The field of 3D digital reconstruction has undergone a soaring and exciting boost with the recent invention of Neural Radiance Fields (NeRF). Within few years researchers have tightened the nuts to improve its photorealistic visual quality from multiple data types and optimized both training and inference time; and thus providing new solutions in several application fields.
Academic Supervisor: Gloria Haro
Supervisor e-mail: gloria.haro@upf.edu
Institution: UPF
Assigned Student Name: Albert Barreiro Díaz
Student e-mail: Albert.Barreiro@autonoma.cat
pre-Assigned Student Name: Albert Barreiro Diaz
Company: Fundacio Eurecat
Contact e-mail: rafael.redondo@eurecat.org
Contact Person: Rafael Redondo
Confidential: No
Date: 2023-04-05 17:25:26
  Utilization of synthetic data for anomaly detection
During the training period, the student will work on Utilization of synthetic data for anomaly detection, which will also be the topic for his master thesis, which will be supervised by Prof. Sergio Escalera. More specifically, the students will get some state-of-the-art anomaly detection algorithm up and running, and then re-train them on a public thermal dataset developed by collaboration with M
Academic Supervisor: Xavier Baro
Supervisor e-mail: xbaro@uoc.edu
Institution: UOC
co-Supervisor: Sergio Escalera
co-Supervisor e-mail: sergio.escalera.guerrero@gmail.com
Assigned Student Name: Miquel Romero Blanch
Student e-mail: Miquel.Romero@autonoma.cat
pre-Assigned Student Name: Miguel Romero Blanch
Company: Milestone Systems Iberia SLU
Contact e-mail: casa@milestone.dk
Contact Person: Kamal Nasrollahi
Confidential: Yes
Date: 2023-03-16 17:54:27
  Semantic Segmentation for Memory-Constrained Edge Devices
The aim of this project is to design a lightweight semantic segmentation network architecture that: (1) is designed to use minimal resources, (2) while maintaining high accuracy, and (3) fits the Midokura production environment. The candidate will work under the mentorship of AI engineers.
Extended abstract: Download PDF
Academic Supervisor: Miguel Angel Gonzalez Ballester
Supervisor e-mail: ma.gonzalez@upf.edu
Institution: UPF
Assigned Student Name: Guillem Martinez Sanchez
Student e-mail: Guillem.MartinezS@autonoma.cat
pre-Assigned Student Name: Guillem Martinez
Company: Midokura
Contact e-mail: magi.toneu@midokura.com
Contact Person: Magi Toneu
Confidential: Yes
Date: 2023-03-15 11:11:41
  Weakly-Supervised Learning for Scene Text Recognition
This project deals with the design and implementation of weakly-supervised learning techniques for the specific case of scene text recognition. Starting from a baseline model, we want to apply weakly-supervised learning techniques to leverage inference data without requiring complex data annotations to incrementally train and enhance the models' performance.
Extended abstract: Download PDF
Academic Supervisor: Dimosthenis Karatzas
Supervisor e-mail: dimos@cvc.uab.cat
Institution: UAB
Assigned Student Name: Kyryl Dubovetskyi
Student e-mail: Kyryl.Dubovetskyi@autonoma.cat
Company: AllRead
Contact e-mail: marcal@allread.ai
Contact Person: Marcal Rossinyol
Confidential: No
Date: 2023-03-08 08:44:31
  Automatic Drift detection and Data Redundancy Removal on Edge Device Deployments
The goal of this internship is to improve a proprietary Automatic Data Control and Collection system. Work in this system focuses on Data Drift Detection, Data Redundancy Removal and Active learning. A PoC will be deployed on edge devices, in order to monitor in an unsupervised way a deployed model. It will also facilitate its retraining in case of performance degradation.
Academic Supervisor: Joost Van de Eijer
Supervisor e-mail: joost@cvc.uab.es
Institution: UAB
Assigned Student Name: Alex Carrillo Alza
Student e-mail: Alex.Carrillo@autonoma.cat
pre-Assigned Student Name: Alex Carrillo Alza
Company: Midokura
Contact e-mail: xavier@midokura.com
Contact Person: Xavier Serra Alza
Confidential: Yes
Date: 2023-02-17 16:12:59
  AI for Earth Observation aboard nano-satellite
The boarding of artificial intelligence (AI) and machine learning (Machine Learning or ML) algorithms on satellites is starting to become a reality. This project explores the algorithms and techniques (mainly Neural-Networks) used for Earth Observation and customizes them to a restricted platform to be aboard a nano-satellite.
Academic Supervisor: Felipe Lumbreras
Supervisor e-mail: felipe.lumbreras@cvc.uab.cat
Institution: UAB
co-Supervisor: Daniel Ponsa
co-Supervisor e-mail: daniel@cvc.uab.cat
Assigned Student Name: Michell Israel Vargas Signoret
Student e-mail: MichellIsrael.Vargas@autonoma.cat
pre-Assigned Student Name: Michell Vargas Signoret
Company: Institut d'Estudis Espacials de Catalunya
Contact e-mail: marius.monton@ieec.cat
Contact Person: Marius Monton
Confidential: No
Date: 2023-02-16 22:06:35
  Generation of synthetic longitudinal magnetic resonance images of subjects with multiple sclerosis.
Multiple sclerosis is a degenerative disease affecting the nervous system. There is no specific test for its diagnosis, but the main tool for its monitoring is magnetic resonance imaging. The aim of this study is to use generative models to train a network that is able to generate images of the patient at different time points from the baseline, thus predicting the development of the disease.
Academic Supervisor: Gerard Marti
Supervisor e-mail: gerard.marti@upf.edu
Institution: UPF
Assigned Student Name: Ana Harris Martínez
Student e-mail: Ana.Harris@autonoma.cat
pre-Assigned Student Name: Ana Harris Martinez
Company: Neuroradiology, Institut de Recerca Vall d’Hebron
Contact e-mail: deborah.pareto.idi@gencat.cat
Contact Person: Deborah Pareto, Neuroradiology Department (IDI)
Confidential: No
Date: 2023-02-06 12:07:09
  LLMs applied to robotic planners
Large language models (LLMs) have unlocked new capabilities of task planning from human instructions. However, grounding such planners to real-world robotic tasks is challenging. On the one hand, robotic capabilities must match with planner outputs, and, on the other hand, the planner must be aware and understand the real-world environment. l
Extended abstract: Download PDF
Academic Supervisor: Bogdan Raducanu
Supervisor e-mail: bogdan@cvc.uab.cat
Institution: UAB
Assigned Student Name: Josep Bravo Bravo
Student e-mail: Josep.Bravo@autonoma.cat
pre-Assigned Student Name: Josep Bravo
Company: Fundacio Eurecat
Contact e-mail: adriana.cruz@eurecat.org
Contact Person: Adriana Cruz
Confidential: No
Date: 2023-01-31 09:43:34
  Improving an algorithm to identify and count books
Our project aims to identify and count the number of books contained in the images to be submitted by respondents to a web survey. An algorithm for these purposes already exists, but it needs to be improved as it presents several problems (no identification of books, or assignment of only one box to two or more books). The aim of the thesis is to overcome these problems by improving the algorithm.
Extended abstract: Download PDF
Academic Supervisor: Dr. Gloria Haro
Supervisor e-mail: gloria.haro@upf.edu
Institution: UPF
Company: IBEI / UPF
Contact e-mail: melanie.revilla@upf.edu
Contact Person: Dr. Melanie Revilla
Confidential: Yes
Date: 2023-01-25 13:27:47
  XAI for anomaly detection
During the training period, the student will work on xai for anomaly detection, which will also be the topic for his master thesis, which will be supervised by Prof. Sergio Escalera. More specifically, the student will get access to some data from Milestone Systems to train some anomaly detection algorithms and utilize existing or developing new xai algorithms to provide an insight into the perfor
Extended abstract: Download PDF
Academic Supervisor: Xavier Baro Sole
Supervisor e-mail: xbaro@uoc.edu
Institution: UOC
co-Supervisor: Sergio Escalera
co-Supervisor e-mail: sergio.escalera.guerrero@gmail.com
Assigned Student Name: Johnny Núñez Cano
Student e-mail: Johnny.Nunez@autonoma.cat
pre-Assigned Student Name: Johnny Nunez Cano
Company: Milestone Systems
Contact e-mail: casa@milestone.dk
Contact Person: Kamal Nasrollahi
Confidential: Yes
Date: 2023-01-17 14:32:27

  2021-2022 proposals
  Self-supervised learning for fish classification
In this project, a self-supervised approach will be searched and explored to train a model to detect and classify fish in underwater images, parting from a small annotated dataset.
Extended abstract: Download PDF
Academic Supervisor: Ramon Baldrich Caselles
Supervisor e-mail: ramon@cvc.uab.cat
Institution: UAB
Assigned Student Name: Alexander Tempelaar Sanchez
Student e-mail: Alexander.Tempelaar@autonoma.cat
pre-Assigned Student Name: Alexander Tempelaar Sanchez
Company: Coronis Computing, SL
Contact e-mail: contact@coronis.es
Contact Person: Rafael Garcia
Confidential: No
Date: 2022-05-31 14:41:50
  Develop an algorithm for multi-person 3D pose estimation from a single camera.
Apply the algorithm to design a way to evaluate some body movements.
Academic Supervisor: Francesc Moreno-Noguer
Supervisor e-mail: francesc.moreno.noguer@upc.edu
Institution: UPC
Assigned Student Name: Adam Szummer Szummer
Student e-mail: Adam.Szummer@autonoma.cat
pre-Assigned Student Name: ADAM SZUMMER
Company: WELLFIT TEC SL
Contact e-mail: andreu@improfit.ai
Contact Person: Andreu Casadella
Confidential: Yes
Date: 2022-05-24 12:04:27
  Automatic dataset update via instance relevance assessment
The aim of this project is to develop a method to automatically add new instances to a dataset according to their relevance. It will involve two main areas: (1) Determine when a dataset has become outdated with respect to the real data being used by an application (2) Determine which new instances are the most relevant to the dataset to optimize data annotation and dataset completeness.
Academic Supervisor: joost van de weijer
Supervisor e-mail: joost@cvc.uab.es
Institution: UAB
Assigned Student Name: Mert Yazan
Student e-mail: Mert.Yazan@autonoma.cat
pre-Assigned Student Name: Mert Yazan
Company: Midokura Iberica S.L
Contact e-mail: xavier@midokura.com
Contact Person: Xavier Serra
Confidential: Yes
Date: 2022-05-16 09:24:20
  Combined Hiper-spectral and RGB imagery for fabric classification
The objective of this project is to develop vision technologies based on artificial intelligence to classify different types of fabrics as a part of its recycling process. In order to differentiate materials with similar appearance, hiper-spectral imagery is a powerful technology. Combining this with high-res RGB images, will improve the segmentation and helps in further recycling steps.
Extended abstract: Download PDF
Academic Supervisor: Ramon Baldrich
Supervisor e-mail: ramon@cvc.uab.cat
Company: Eurecat - Centre Tecnologic de Catalunya
Contact e-mail: ferran.roure@eurecat.org
Contact Person: Ferran Roure
Confidential: No
Date: 2022-05-13 11:23:24
  Photorealistic Facial Wrinkle Removal
Recently, Augmented Reality has been used to simulate skin treatments on the facial region. However, the obtained results are not photorealistic. In this MA dissertation, we will study deep learning techniques that allow users to simulate skin treatments in a photorealistic manner. In concrete, we will focus on the problem of photorealistic wrinkles removal.
Extended abstract: Download PDF
Academic Supervisor: Coloma Ballester
Supervisor e-mail: coloma.ballester@upf.edu
Institution: UPF
Assigned Student Name: Marcelo Sanchez Ortega
Student e-mail: Marcelo.Sanchez@autonoma.cat
pre-Assigned Student Name: Marcelo Sanchez
Company: Crisalix
Contact e-mail: eduard.ramon@crisalix.com
Contact Person: Eduard Ramon
Confidential: Yes
Date: 2022-05-09 18:08:11
  Structured light in high dose and underwater environment
The goal of the project is to produce a point cloud using structured light (SL) methods in an underwater and high dose radiation environment. The SL is created with a laser and a glass mask which has mathematical encoding uniquely identifying the laser row number with 10 bands. The laser camera pair pans 360 degree and captures how the SL is displaced. We aim to obtain depth based 10 laser bands.
Academic Supervisor: Josep Ramon Casas
Supervisor e-mail: josep.ramon.casas@upc.edu
Institution: UPC
Assigned Student Name: Daniel Rodriguez Estevez
Student e-mail: Daniel.RodriguezEs@autonoma.cat
pre-Assigned Student Name: Daniel Rodriguez Estevez
Company: Createc
Contact e-mail: ahmet.cinar@createc.co.uk
Contact Person: Ahmet Cinar
Confidential: Yes
Date: 2022-05-04 19:52:39
  Neural fields for multi-view 3D reconstruction using noisy camera poses
Recent approaches for multi-view 3D reconstruction achieve good results by using learned representations for the geometry and appearance of the object. One limitation of these methods is the requirement of good camera poses. In this project we investigate new self-calibrating 3D reconstruction methods using neural fields that can work with a noisy set of camera pose initialisations.
Academic Supervisor: Francesc Moreno
Supervisor e-mail: francesc.moreno.noguer@upc.edu
Institution: UPC
co-Supervisor: Josep Ramon Casas
co-Supervisor e-mail: josep.ramon.casas@upc.edu
Assigned Student Name: Ibrar Malik Ara
Student e-mail: Ibrar.Malik@autonoma.cat
pre-Assigned Student Name: Ibrar Malik Ara
Company: Crisalix Labs SLU
Contact e-mail: admin@crisalix.com
Contact Person: Patricia Castro
Confidential: Yes
Date: 2022-05-04 11:36:58
  State of the art technology OCR system
Train a general free and open OCR model for the Latin-Spanish alphabet capable of competing in accuracy with Google Vision API and Amazon Textract. PaddleOCR uses the latest refinement techniques for learning OCR lightweight neural networks designed for the IoT domain.
Extended abstract: Download PDF
Academic Supervisor: Ernest Valveny
Supervisor e-mail: ernest@cvc.uab.cat
Institution: UAB
Company: Rosepetal SL
Contact e-mail: info@rosepetal.ai
Contact Person: Pere Gironès
Confidential: Yes
Date: 2022-04-29 18:04:52
  Lightweight neural network architectures for on-device object detection
The aim of this project is to design a lightweight detection network architecture that: (a) allows usage of higher resolution input, and (b) is sufficiently small to fit into edge devices. The candidate will work under the mentorship of AI engineers. The resulting architecture would be deployed on smart sensor devices.
Extended abstract: Download PDF
Academic Supervisor: Maria Vanrell
Supervisor e-mail: maria.vanrell@uab.cat
Institution: UAB
Assigned Student Name: Alex Martin Martinez
Student e-mail: Alex.MartinM@autonoma.cat
pre-Assigned Student Name: Alex Martin Martinez
Company: Midokura Iberica
Contact e-mail: ekaterina@midokura.com
Contact Person: Ekaterina Kravchenko
Confidential: Yes
Date: 2022-04-25 14:21:13
  Computer vision algorithm application for insect rearing optimization
Through object recognition and tracking, this project aims to understand and appraise larval population behavior per container in real-time. The main goal is to monitor the larval growth and maturity stage from the visible individuals in the surface and assess the need of additional feed (identifying degree of substrate exhaustion) for each container.
Extended abstract: Download PDF
Academic Supervisor: Antonio Agudo
Supervisor e-mail: antonio.agudo@upc.edu
Company: InsFeed
Contact e-mail: marta.campos@insfeed.com
Contact Person: Marta Campos
Confidential: Yes
Date: 2022-03-04 11:35:36
  Automated Video Edition Using Representations From Collaborative Experts
We will investigate methods enabling Automated Video Edition (AVE). The AVE system will select which among a set of synchronized video shots should be selected, in order to create a video of a pre-defined duration that combines the contributions of different shots. The work will build on top of an existing attention-based automated video editor and explore the use of collaborative experts.
Extended abstract: Download PDF
Academic Supervisor: Lluís Gómez
Supervisor e-mail: lgomez@cvc.uab.cat
Institution: UAB
Company: La Rumba de Barcelona S.L.
Contact e-mail: joan.llobera@izirecord.com
Contact Person: Joan Llobera
Confidential: Yes
Date: 2022-03-01 02:18:49
  Sports Biomechanical Analysis
The aim of this project is to develop an automatic bio-mechanical analysis solution applied to padel that by recording yourself when practicing is able to examine your movements in real time and show you where you need to improve. The candidate will work very closely to experts in the field of padel and computer vision to develop and implement the algorithm that will be used by a professionals
Extended abstract: Download PDF
Academic Supervisor: Antonio Agudo
Supervisor e-mail: antonio.agudo@upc.edu
Institution: UPC
Company: Lob Tech SL.
Contact e-mail: oscar.lopez@lob.cat
Contact Person: Oscar Lopez
Confidential: Yes
Date: 2022-02-23 15:38:32
  Lane Detection for Automated Driving, a Reference Perception System.
Development of algorithms for lane detection and tracking from images and point clouds. The algorithms should combine information from different sensors and timestamps to achieve more robust results. The system will be integrated in a reference perception system that will be used to evaluate lane detection system and related ADAS functions (LDW, ALK, ...) on commercial vehicles
Extended abstract: Download PDF
Academic Supervisor: Antonio Agudo
Supervisor e-mail: antonio.agudo@upc.edu
Institution: UPC
Assigned Student Name: Sergi Vidal Bazan
Student e-mail: Sergi.VidalB@autonoma.cat
Company: IDIADA Automotive Technology, S.A.
Contact e-mail: monica.lores@idiada.com
Contact Person: Monica Lores Garcia
Confidential: Yes
Date: 2022-02-22 12:08:03
  Lightweight Monocular 3D Vehicle Detection
This project deals with the design and training of an object detection model able to produce 3D bounding-boxes for different types of vehicles in monocular images. Taking inspiration from the CenterNet model, we want to obtain a lightweight 3D object detection model that can run in real time in low-resource environments.
Extended abstract: Download PDF
Academic Supervisor: Dimosthenis Karatzas
Supervisor e-mail: dimos@cvc.uab.es
Institution: UAB
Assigned Student Name: Eric Henriksson Martí
Student e-mail: Eric.Henriksson@autonoma.cat
Company: AllRead
Contact e-mail: marcal@allread.ai
Contact Person: Marçal Rossinyol
Confidential: Yes
Date: 2022-02-16 12:20:17
  Neural point features for automatic wind turbine inspection
Blade defects are detected through images captured in wind turbine inspections, so they can be repaired in due time. This project focuses on obtaining meaningful deep-based image features that can be easily distinguishable between them, so we can use them to stitch the images and create a panoramic/mosaic or 3D model, just to name a few. The project will be done in IRI-UPC (Barcelona).
Extended abstract: Download PDF
Academic Supervisor: Antonio Agudo
Supervisor e-mail: antonio.agudo@upc.edu
Institution: UPF
Company: Wind Power LAB
Contact e-mail: rpg@windpowerlab.com
Contact Person: Raül Pérez
Confidential: Yes
Date: 2022-02-15 13:29:37
  Identification of product attributes by computer vision
The project consists of identifying product attributes from photos. We have a dataset of product photos with associated metadata (packaging, brand, barcode, ...) and we want to automate the identification of these attributes for the photos of new products entering the system.
Academic Supervisor: Coloma Ballester
Supervisor e-mail: coloma.ballester@upf.edu
Institution: UPF
co-Supervisor: Gloria Haro
co-Supervisor e-mail: gloria.haro@upf.edu
Company: Sagaci Research
Contact e-mail: hr@sagaciresearch.com
Contact Person: Joao Terlica
Confidential: Yes
Date: 2022-02-04 14:40:15