Unai Elordi

Dr. Unai Elordi

orcid

Research Focus & Metrics

Academic Career Length

15Years (2012-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
9 October 2020 in IEEE Software
Verified
Journal Article
Closed access
UEUnai Elordi
Unai Elordi
LULuis Unzueta
Luis Unzueta
JGJon Goenetxea
Jon Goenetxea
Sergio Sanchez-CarballidoSergio Sanchez-Carballido
Sergio Sanchez-Carballido
IAIgnacio Arganda-Carreras
Ignacio Arganda-Carreras
OOOihana Otaegui
Oihana Otaegui
Abstract—Current computing requirements for high-scale inference of Deep Neural Networks (DNNs) demand distributed execution environments. The advantages of serverless functions in distributed computation offloading and automatic resource scalability make them a very suitable environment for such a task. However, finding the optimal workload at minimum resource usage requires extensive benchmarking analysis such as those of the MLPerf standard. However, due to its monolithic nature, the MLPerf execution model collides with serverless function platforms. To address this issue, we provide a novel decomposition methodology from the current MLPerf benchmark to the serverless function execution model. Moreover, we have tested our approach in Amazon Lambda to benchmark the processing capabilities of OpenCV and OpenVINO IE DNN inference engines using Caffe, Tensorflow, and OpenVINO models for Computer Vision tasks. Experimental results prove Amazon Lambda a suitable platform for DNN inference following our proposed architecture.

1 May 2014 in Journal of Real-Time Image Processing
Verified
Journal Article
Closed access
PDF file
LDLeonardo De-Maeztu
Leonardo De-Maeztu
UEUnai Elordi
Unai Elordi
Marcos NietoMarcos Nieto
Marcos Nieto
JBJavier Barandiaran
Javier Barandiaran
OOOihana Otaegui
Oihana Otaegui

14 April 2023 in IEEE Journal of Selected Topics in Signal Processing
Verified
Journal Article
Closed access
ARAndrea Macarulla Rodriguez
Andrea Macarulla Rodriguez
LULuis Unzueta
Luis Unzueta
Zeno GeradtsZeno Geradts
Zeno Geradts
MWMarcel Worring
Marcel Worring
UEUnai Elordi
Unai Elordi
Identifying suspects from surveillance footage is a crucial task in forensic investigations, but it is often hindered by the variable conditions of observation and the large amounts of data. Face image quality (FIQ) is a metric that measures the usefulness of a face sample for facial recognition. Existing methods for automated FIQ assessment only provide a scalar value for quality, and do not indicate which factors are causing low quality. Additionally, these methods are computationally expensive, which makes current FIQ assessment methods unsuitable for large numbers of images. To address these issues, we introduce multi-task explainable quality networks (XQNets). XQNets provide both the quality value and the associated facial and environmental attributes, and automatically learn how each attribute contributes to the quality value during the training process. We also propose a dataset-agnostic quality pairing protocol to ensure that sample pairs are balanced across datasets and evaluations are fair. Our experiments on the LFW and SCface benchmarks show that our approach generalizes well across different datasets and outperforms state-of-the-art methods. Our method offers a fast, explainable approach to FIQ assessment, making it suitable for large-scale forensic applications.

20 December 2021 in Information
Verified
Journal Article
Open access
PDF file
UEUnai Elordi
Unai Elordi
CLChiara Lunerti
Chiara Lunerti
LULuis Unzueta
Luis Unzueta
JGJon Goenetxea
Jon Goenetxea
NANerea Aranajuelo
Nerea Aranajuelo
IAIgnacio Arganda-Carreras
Ignacio Arganda-Carreras
In this paper, we tackle the problem of deploying face recognition (FR) solutions in heterogeneous Internet of Things (IoT) platforms. The main challenges are the optimal deployment of deep neural networks (DNNs) in the high variety of IoT devices (e.g., robots, tablets, smartphones, etc.), the secure management of biometric data while respecting the users’ privacy, and the design of appropriate user interaction with facial verification mechanisms for all kinds of users. We analyze different approaches to solving all these challenges and propose a knowledge-driven methodology for the automated deployment of DNN-based FR solutions in IoT devices, with the secure management of biometric data, and real-time feedback for improved interaction. We provide some practical examples and experimental results with state-of-the-art DNNs for FR in Intel’s and NVIDIA’s hardware platforms as IoT devices.

17 June 2018
Verified
Book Chapter
Closed access
PDF file
UEUnai Elordi
Unai Elordi
LULuis Unzueta
Luis Unzueta
IAIgnacio Arganda-Carreras
Ignacio Arganda-Carreras
OOOihana Otaegui
Oihana Otaegui

28 August 2023 in IEEE Access
Verified
Journal Article
Open access
UEUnai Elordi
Unai Elordi
Nerea AranjueloNerea Aranjuelo
Nerea Aranjuelo
LULuis Unzueta
Luis Unzueta
JAJose Luis Apellaniz
Jose Luis Apellaniz
IAIgnacio Arganda-Carreras
Ignacio Arganda-Carreras
In this paper, we propose an approach to optimize the deployment of on-board video analytics for checking the correct positioning of luggage in aircraft cabins. The system consists of embedded cameras installed on top of the cabin and a heterogeneous embedded processor. Each camera covers multiple regions of interest (i.e., multiple seats or aisle sections) to minimize the number of cameras required. Each image region is processed by a separate image classification algorithm trained with the expected kind of visual appearance considering the effect of perspective and lens distortion. They classify these regions as correct or incorrect for cabin readiness by exploiting the hierarchical structure of classes composed of different configurations of passengers’ and objects’ presence or absence and the objects’ location. Our approach leverages semantic distances between classes to guide prototypical neural networks for multi-tasking between the main classification (i.e., correct or incorrect status) and auxiliary attributes (i.e., scene configurations), learning robust features from different data domains (i.e., various cabins, real or synthetic). The processing pipeline optimizes response delay and power consumption by leveraging embedded processors’ computing capabilities. We carried out experiments in a cabin mockup with a Jetson AGX Xavier, efficiently obtaining better-quality descriptive information from the scene to improve the system’s accuracy compared to alternative state-of-the-art methods.