Skip to main content

APRENDIZAJE AUTOMATICO BIOINSPIRADO

PID2020-118071GB-I00

•
Funding agency name Agencia Estatal de Investigación
Funding agency acronym AEI
Program Programa Estatal de I+D+i Orientado a los Retos de la Sociedad
Call Proyectos de I+D+i (Generación de Conocimiento y Retos Investigación)
Call year 2020
State Plan 2017-2020 Plan Estatal de Investigación Científica y Técnica y de Innovación
Beneficiary centre UNIVERSITAT DE VALÈNCIA

Resultados relacionados

Resultados totales (Incluyendo duplicados): 15
Encontrada(s) 1 página(s)

Repositorio Digital de la UPF
Publication Artículo científico (JournalArticle)

Contrast sensitivity functions in autoencoders

  • Li, Qiang
  • Gomez-Villa, Alex
  • Bertalmío, Marcelo
  • Malo, Jesús
Three decades ago, Atick et al. suggested that human frequency sensitivity may emerge from the enhancement required for a more efficient analysis of retinal images. Here we reassess the relevance of low-level vision tasks in the explanation of the contrast sensitivity functions (CSFs) in light of 1) the current trend of using artificial neural networks for studying vision, and 2) the current knowledge of retinal image representations. As a first contribution, we show that a very popular type of convolutional neural networks (CNNs), called autoencoders, may develop human-like CSFs in the spatiotemporal and chromatic dimensions when trained to perform some basic low-level vision tasks (like retinal noise and optical blur removal), but not others (like chromatic) adaptation or pure reconstruction after simple bottlenecks). As an illustrative example, the best CNN (in the considered set of simple architectures for enhancement of the retinal signal) reproduces the CSFs with a root mean square error of 11% of the maximum sensitivity. As a second contribution, we provide experimental evidence of the fact that, for some functional goals (at low abstraction level), deeper CNNs that are better in reaching the quantitative goal are actually worse in replicating human-like phenomena (such as the CSFs). This low-level result (for the explored networks) is not necessarily in contradiction with other works that report advantages of deeper nets in modeling higher level vision goals. However, in line with a growing body of literature, our results suggests another word of caution about CNNs in vision science because the use of simplified units or unrealistic architectures in goal optimization may be a limitation for the modeling and understanding of human vision., Partially funded by these grants from GVA/AEI/FEDER/EU: MICINN DPI2017- 89867-C2-2-R, MICINN PID2020-118071GB-I00, and GVA Grisolía-P/2019/035 (for JM and QL), and MICINN PGC2018-099651-B-I00 (for A.G.V. and M.B.).
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

Elementary Clothoid-Based Three-Dimensional Curve for Unmanned Aerial Vehicles

RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

Combining haptics and inertial motion capture to enhance remote control of a dual-arm robot

  • Girbés-Juan, Vicent
  • Schettino, Vinicius
  • Demiris, Yiannis
  • Gracia Calandin, Luis Ignacio|||0000-0001-9258-9286
  • Solanes, J. Ernesto|||0000-0001-5011-4872
  • Tornero Montserrat, Josep|||0000-0003-1571-0251
[EN] High dexterity is required in tasks in which there is contact between objects, such as surface conditioning (wiping, polishing, scuffing, sanding, etc.), specially when the location of the objects involved is unknown or highly inaccurate because they are moving, like a car body in automotive industry lines. These applications require the human adaptability and the robot accuracy. However, sharing the same workspace is not possible in most cases due to safety issues. Hence, a multi-modal teleoperation system combining haptics and an inertial motion capture system is introduced in this work. The human operator gets the sense of touch thanks to haptic feedback, whereas using the motion capture device allows more naturalistic movements. Visual feedback assistance is also introduced to enhance immersion. A Baxter dual-arm robot is used to offer more flexibility and manoeuvrability, allowing to perform two independent operations simultaneously. Several tests have been carried out to assess the proposed system. As it is shown by the experimental results, the task duration is reduced and the overall performance improves thanks to the proposed teleoperation method., This research was funded by Generalitat Valenciana (Grants GV/2021/074 and GV/2021/181) and by the SpanishGovernment (Grants PID2020-118071GB-I00 and PID2020-117421RBC21 funded by MCIN/AEI/10.13039/501100011033). This work was also supported byCoordenacao de Aperfeiaoamento de Pessoal de Nivel Superior (CAPES Brasil) under Finance Code 001, by CEFET-MG, and by a Royal Academy of Engineering Chair in Emerging Technologies to YD.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

Cautious Bayesian Optimization: A Line Tracker Case Study

  • Girbés-Juan, Vicent
  • Moll, Joaquín
  • Sala, Antonio|||0000-0002-5691-8772
  • Leopoldo Armesto|||0000-0003-0979-4428
[EN] In this paper, a procedure for experimental optimization under safety constraints, to be denoted as constraint-aware Bayesian Optimization, is presented. The basic ingredients are a performance objective function and a constraint function; both of them will be modeled as Gaussian processes. We incorporate a prior model (transfer learning) used for the mean of the Gaussian processes, a semi-parametric Kernel, and acquisition function optimization under chance-constrained requirements. In this way, experimental fine-tuning of a performance objective under experiment-model mismatch can be safely carried out. The methodology is illustrated in a case study on a line-follower application in a CoppeliaSim environment., This research was funded by MCIN/AEI/10.13039/501100011033, Agencia Estatal de Investigación (Spanish government), grant numbers PID2020-116585GB-I00 and PID2020-118071GB-I00.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

Asynchronous Sensor Fusion of GPS, IMU and CAN-Based Odometry for Heavy-Duty Vehicles

  • Girbes-Juan, Vicent
  • Hernandez-Ferrandiz, Daniel
  • Leopoldo Armesto|||0000-0003-0979-4428
  • Dols Ruiz, Juan Francisco|||0000-0003-1815-1360
  • Sala, Antonio|||0000-0002-5691-8772
[EN] In heavy-duty vehicles, multiple signals are available to estimate the vehicle's kinematics, such as Inertial Measurement Unit (IMU), Global Positioning System (GPS) and linear and angular speed readings from wheel tachometers on the internal Controller Area Network (CAN). These signals have different noise variance, bandwidth and sampling rate (being the latter, possibly, irregular). In this paper we present a non-linear sensor fusion algorithm allowing asynchronous sampling and non-causal smoothing. It is applied to achieve accuracy improvements when incorporating odometry measurements from CAN bus to standard GPS+IMU kinematic estimation, as well as the robustness against missing data. Our results show that this asynchronous multi-sensor (GPS+IMU+CAN-based odometry) fusion is advantageous in low-speed manoeuvres, improving accuracy and robustness to missing data, thanks to non-causal filtering. The proposed algorithm is based on Extended Kalman Filter and Smoother, with exponential discretization of continuous-time stochastic differential equations, in order to process measurements at arbitrary time instants; it can provide data to subsequent processing steps at arbitrary time instants, not necessarily coincident with the original measurement ones. Given the extra information available in the smoothing case, its estimation performance is less sensitive to the noise-variance parameter setting, compared to causal filtering. Working Matlab code is provided at the end of this work., This research was supported in part by the Agencia Espanola de Investigacion (European Union) under Grants PID2020-116585GB-I00 and PID2020-118071GB-I00, and in part by the Generalitat Valenciana under Grant GV/2021/074. The review of this article was coordinated by Dr. Sohel Anwar. (Corresponding author: Vicent Girbes-Juan.)
RUA. Repositorio Institucional de la Universidad de Alicante
Publication Otros (Other)

What is beautiful is still good: the attractiveness halo effect in the era of beauty filters

  • Gulati, Aditya
  • Martínez-Garcia, Marina
  • Fernández, Daniel
  • Lozano, Miguel Angel
  • Lepri, Bruno
  • Oliver, Nuria
  • Lozano, Miguel Angel|||/items/413f111d-7a63-43f2-bb70-6421affe1647
The impact of cognitive biases on decision-making in the digital world remains under-explored despite its well-documented effects in physical contexts. This paper addresses this gap by investigating the attractiveness halo effect using AI-based beauty filters. We conduct a large-scale online user study involving 2748 participants who rated facial images from a diverse set of 462 distinct individuals in two conditions: original and attractive after applying a beauty filter. Our study reveals that the same individuals receive statistically significantly higher ratings of attractiveness and other traits, such as intelligence and trustworthiness, in the attractive condition. We also study the impact of age, gender and ethnicity and identify a weakening of the halo effect in the beautified condition, resolving conflicting findings from the literature and suggesting that filters could mitigate this cognitive bias. Finally, our findings raise ethical concerns regarding the use of beauty filters.
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publication Artículo científico (JournalArticle)

Contrast sensitivity functions in autoencoders

  • Li, Qiang
  • Gómez-Villa, Alex
  • Bertalmío, Marcelo
  • Malo, Jesús
45 pags., 24 figs., 4 tabs., 3 apps., The human contrast sensitivity function (CSF) characterizes the psychophysical response to visual gratings of different frequency (Campbell & Robson, 1968). Filter characterizations in the Fourier domain are complete only for linear, shift-invariant systems. Human vision certainly is more complicated than that, however, this simple measure of the bandwidth of the system is still of paramount significance in biological vision: the CSF filter is an image-computable model that roughly describes the kind of visual information that is available for humans (Watson & Ahumada, 2016). Moreover, although it is defined for threshold conditions, there are many examples that illustrate the relevance of the CSF in more general situations (Watson et al., 1986; Watson & Malo, 2002; Watson & Ahumada, 2005), so it has shaped image engineering over decades (Mannos & Sakrison, 1974; Hunt, 1975; Wallace, 1992; Taubman & Marcellin, 2001). This theoretical and practical relevance motivated the measurement of CSFs, not only for spatial gratings (Campbell & Robson, 1968), but also for moving gratings (Kelly, 1979), chromatic gratings (Mullen, 1985), spatiotemporal chromatic gratings (Díez-Ajenjo et al., 2011), at different luminance levels (Wuerger et al., 2020), and for alternative basis of the image space (Malo et al., 1997), Partially funded by these grants from GVA/AEI/FEDER/EU: MICINN DPI2017-89867-C2-2-R, MICINN PID2020-118071GB-I00, and GVA Grisolía-P/2019/035 (for JM and QL), and MICINN PGC2018-099651-B-I00 (for A.G.V. and M.B.)
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publication Artículo científico (JournalArticle)

Spectral Pre-Adaptation for Restoring Real-World Blurred Images using Standard Deconvolution Methods

  • Dong, Chaoqun
  • Sroubek, F.
  • Portilla, Javier
29 pags., 13 figs., 2 tabs., Classical blur models are based on simplifying assumptions, namely shift-equivariance and cir-cular boundary condition (CBC), that rarely hold in practice. Shift-equivariance means that a shift of the input produces the same shift of the output, which implies that blur is spatially invariant and image aliasing is not present. The CBC assumes that the image is rectangular and periodically repeating. Discrepancies between simplified models and real blurred observations cause strong artifacts in image restoration. The common remedy is to increase the model complexity and remove simplifying assumptions. However, this also brings extra computational complexity to the restoration task. We present spectral pre-adaptation (SPA) that pre-processes blurred images so they can be restored using fast standard deconvolution algorithms suitable for simplified models. The SPA serves as a connector between classical deconvolution methods and a variety of real observations involving blur. Experiments on simulated and real images show that standard deconvolution of SPA-interpolated images not only greatly reduces artifacts compared to direct deconvolution, but performs on a par with more complex restoration methods., This work has been funded by the Spanish Government grants FIS2016-75891-P, PID2020-118071GBI00 and PID2020-113596GB-I00, the Czech Science Foundation grant GA20-27939S and the Praemium Academiae awarded by the Czech Academy of Sciences
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publication Artículo científico (JournalArticle)

Using Decoupled Features for Photorealistic Style Transfer

  • Canham, Trevor
  • Martín-Fernández, Adrián
  • Bertalmio, Marcelo
  • Portilla, Javier
45 pags., 23 figs., 8 algorithms, 1 tab. -- AMS subject classifications. 68Q25, 68R10, 68U05, In this work we propose a photo-realistic style transfer method for image and video that is based on vision science principles and on a recent mathematical formulation for the deterministic decoupling of sample statistics. The novel aspects of our approach include matching decoupled moments of higher order than in common style transfer approaches, and matching a descriptor of the power spectrum so as to characterize and transfer diffusion effects between source and target, which is something that has not been considered before in the literature. The results are of high visual quality, without spatio-temporal artifacts, and validation tests in the form of observer preference experiments show that our method compares very well with the state-of-the-art. The computational complexity of the algorithm is low, and we propose a numerical implementation that is amenable for real-time video application. Finally, another contribution of our work is to point out that current deep learning approaches for photo-realistic style transfer don’t really achieve photo-realistic quality outside of limited examples, because the results too often show unacceptable visual artifacts., This work has received funding from the Spanish Government grants PID2020-118071GB-I00, PID2020-113596GB-I00 and PID2021-127373NB-I00, and from the European Union’s Horizon 2020 research and innovation program under grant agreement number 952027 (project AdMiRe)., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publication Artículo científico (JournalArticle)

Cortical Divisive Normalization from Wilson–Cowan Neural Dynamics

  • Malo, Jesús
  • Esteve-Taboada, J.J.
  • Bertalmio, Marcelo
39 pages, 14 figures, 1 table, Divisive Normalization and the Wilson–Cowan equations are well-known influential models of nonlinear neural interaction (Carandini and Heeger in Nat Rev Neurosci 13(1):51, 2012; Wilson and Cowan in Kybernetik 13(2):55, 1973). However, they have been always treated as different approaches and have not been analytically related yet. In this work, we show that Divisive Normalization can be derived from the Wilson–Cowan dynamics. Specifically, assuming that Divisive Normalization is the steady state of the Wilson–Cowan differential equations, we find that the kernel that controls neural interactions in Divisive Normalization depends on the Wilson–Cowan kernel but also depends on the signal. A standard stability analysis of a Wilson–Cowan model with the parameters obtained from our relation shows that the Divisive Normalization solution is a stable node. This stability suggests the appropriateness of our steady state assumption. The proposed theory provides a mechanistic foundation for the suggestions that have been done on the need of signal-dependent Divisive Normalization in Coen-Cagli et al. (PLoS Comput Biol 8(3):e1002405, 2012). Moreover, this theory explains the modifications that had to be introduced ad hoc in Gaussian kernels of Divisive Normalization in Martinez-Garcia et al. (Front Neurosci 13:8, 2019) to reproduce contrast responses in V1 cortex. Finally, the derived relation implies that the Wilson–Cowan dynamics also reproduce visual masking and subjective image distortion, which up to now had been explained mainly via Divisive Normalization., This work was partially funded by the Spanish Ministerio de Ciencia e Innovación (MICIIN/FEDER/UE) projects PID2020-118071GB-I00 and PDC2021-121522-C21 and by the Generalitat Valenciana Grants GrisoliaP/2019/035 and CIPROM/2021/056. Open Access funding provided thanks to the CRUE-CSIC agreement with Springer Nature., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publication Artículo científico (JournalArticle)

Cortical Divisive Normalization from Wilson–Cowan Neural Dynamics

  • Malo, Jesús
  • Esteve-Taboada, J.J.
  • Bertalmio, Marcelo
39 pages, 14 figures, 1 table, Divisive Normalization and the Wilson–Cowan equations are well-known influential models of nonlinear neural interaction (Carandini and Heeger in Nat Rev Neurosci 13(1):51, 2012; Wilson and Cowan in Kybernetik 13(2):55, 1973). However, they have been always treated as different approaches and have not been analytically related yet. In this work, we show that Divisive Normalization can be derived from the Wilson–Cowan dynamics. Specifically, assuming that Divisive Normalization is the steady state of the Wilson–Cowan differential equations, we find that the kernel that controls neural interactions in Divisive Normalization depends on the Wilson–Cowan kernel but also depends on the signal. A standard stability analysis of a Wilson–Cowan model with the parameters obtained from our relation shows that the Divisive Normalization solution is a stable node. This stability suggests the appropriateness of our steady state assumption. The proposed theory provides a mechanistic foundation for the suggestions that have been done on the need of signal-dependent Divisive Normalization in Coen-Cagli et al. (PLoS Comput Biol 8(3):e1002405, 2012). Moreover, this theory explains the modifications that had to be introduced ad hoc in Gaussian kernels of Divisive Normalization in Martinez-Garcia et al. (Front Neurosci 13:8, 2019) to reproduce contrast responses in V1 cortex. Finally, the derived relation implies that the Wilson–Cowan dynamics also reproduce visual masking and subjective image distortion, which up to now had been explained mainly via Divisive Normalization., This work was partially funded by the Spanish Ministerio de Ciencia e Innovación (MICIIN/FEDER/UE) projects PID2020-118071GB-I00 and PDC2021-121522-C21 and by the Generalitat Valenciana Grants GrisoliaP/2019/035 and CIPROM/2021/056. Open Access funding provided thanks to the CRUE-CSIC agreement with Springer Nature.
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publication Artículo científico (JournalArticle)

Evaluating Perceptual Distance Models by Fitting Binomial Distributions to Two-Alternative Forced Choice Data

  • Hepburn, Alex
  • Santos-Rodríguez, Raúl
  • Portilla, Javier
19 pages, 10 figures, 8 tables, The Two-Alternative Forced-Choice (2AFC) paradigm offers advantages over the Mean Opinion Score (MOS) paradigm in psychophysics (PF), such as simplicity and robustness. However, when evaluating perceptual distance models, MOS enables direct correlation between model predictions and PF data. In contrast, 2AFC only allows pairwise comparisons to be converted into a quality ranking similar to MOS when comparisons include shared images. In large datasets, like BAPPS, where image patches and distortions are combined randomly, deriving rankings from 2AFC PF data becomes infeasible, as distorted images included in each comparison are independent. To address this, instead of relying on MOS correlation, researchers have trained ad-hoc neural networks to reproduce 2AFC PF data based on pairs of model distances -- a black-box approach with conceptual and operational limitations. This paper introduces a more robust distance-model evaluation method using a pure probabilistic approach, applying maximum likelihood estimation to a binomial decision model. Our method demonstrates superior simplicity, interpretability, flexibility, and computational efficiency, as shown through evaluations of various visual distance models on two 2AFC PF datasets., This work was funded by UKRI Turing AI Fellowship EP/V024817/1, TAILOR, a project funded by the EU Horizon 2020 research and innovation programme under grant agreement no. 952215, by the Spanish Government grants PID2020-118071GB-I00 and PID2020 113596GB-I00, the VIS4NN project- Programa Fundamentos de la Fundación BBVA 2022, and by the TEC-2024/COM-322 (IDEAL-CV-CM) project from the Comunidad de Madrid.
Recercat. Dipósit de la Recerca de Catalunya
Publication Artículo científico (JournalArticle)

Contrast sensitivity functions in autoencoders

  • Li, Qiang
  • Gomez-Villa, Alex
  • Bertalmío, Marcelo
  • Malo, Jesús
Three decades ago, Atick et al. suggested that human frequency sensitivity may emerge from the enhancement required for a more efficient analysis of retinal images. Here we reassess the relevance of low-level vision tasks in the explanation of the contrast sensitivity functions (CSFs) in light of 1) the current trend of using artificial neural networks for studying vision, and 2) the current knowledge of retinal image representations. As a first contribution, we show that a very popular type of convolutional neural networks (CNNs), called autoencoders, may develop human-like CSFs in the spatiotemporal and chromatic dimensions when trained to perform some basic low-level vision tasks (like retinal noise and optical blur removal), but not others (like chromatic) adaptation or pure reconstruction after simple bottlenecks). As an illustrative example, the best CNN (in the considered set of simple architectures for enhancement of the retinal signal) reproduces the CSFs with a root mean square error of 11% of the maximum sensitivity. As a second contribution, we provide experimental evidence of the fact that, for some functional goals (at low abstraction level), deeper CNNs that are better in reaching the quantitative goal are actually worse in replicating human-like phenomena (such as the CSFs). This low-level result (for the explored networks) is not necessarily in contradiction with other works that report advantages of deeper nets in modeling higher level vision goals. However, in line with a growing body of literature, our results suggests another word of caution about CNNs in vision science because the use of simplified units or unrealistic architectures in goal optimization may be a limitation for the modeling and understanding of human vision., Partially funded by these grants from GVA/AEI/FEDER/EU: MICINN DPI2017- 89867-C2-2-R, MICINN PID2020-118071GB-I00, and GVA Grisolía-P/2019/035 (for JM and QL), and MICINN PGC2018-099651-B-I00 (for A.G.V. and M.B.).
Dipòsit Digital de Documents de la UAB
Publication Artículo científico (JournalArticle)

Contrast sensitivity functions in autoencoders

  • Li, Qiang
  • Gómez Villa, Alexandra|||0000-0003-0469-3425
  • Bertalmío, Marcelo|||0000-0002-1023-8325
  • Malo, Jesús|||0000-0002-5684-8591
Three decades ago, Atick et al. suggested that human frequency sensitivity may emerge from the enhancement required for a more efficient analysis of retinal images. Here we reassess the relevance of low-level vision tasks in the explanation of the contrast sensitivity functions (CSFs) in light of 1) the current trend of using artificial neural networks for studying vision, and 2) the current knowledge of retinal image representations. As a first contribution, we show that a very popular type of convolutional neural networks (CNNs), called autoencoders, may develop human-like CSFs in the spatiotemporal and chromatic dimensions when trained to perform some basic low-level vision tasks (like retinal noise and optical blur removal), but not others (like chromatic) adaptation or pure reconstruction after simple bottlenecks). As an illustrative example, the best CNN (in the considered set of simple architectures for enhancement of the retinal signal) reproduces the CSFs with a root mean square error of 11% of the maximum sensitivity. As a second contribution, we provide experimental evidence of the fact that, for some functional goals (at low abstraction level), deeper CNNs that are better in reaching the quantitative goal are actually worse in replicating human-like phenomena (such as the CSFs). This low-level result (for the explored networks) is not necessarily in contradiction with other works that report advantages of deeper nets in modeling higher level vision goals. However, in line with a growing body of literature, our results suggests another word of caution about CNNs in vision science because the use of simplified units or unrealistic architectures in goal optimization may be a limitation for the modeling and understanding of human vision.
Dipòsit Digital de Documents de la UAB
Publication Artículo científico (JournalArticle)

A Scalable Reduced-Complexity Compression of Hyperspectral Remote Sensing Images Using Deep Learning

  • Mijares i Verdú, Sebastià|||0000-0003-1038-6413
  • Ballé, Johannes|||0000-0003-0769-8985
  • Laparra, Valero|||0000-0001-7531-9890
  • Bartrina-Rapesta, Joan|||0000-0002-1551-3680
  • Hernández-Cabronero, Miguel|||0000-0001-9301-4337
  • Serra-Sagristà, Joan|||0000-0003-4729-9292
Two key hurdles to the adoption of Machine Learning (ML) techniques in hyperspectral data compression are computational complexity and scalability for large numbers of bands. These are due to the limited computing capacity available in remote sensing platforms and the high computational cost of compression algorithms for hyperspectral data, especially when the number of bands is large. To address these issues, a channel clusterisation strategy is proposed, which reduces the computational demands of learned compression methods for real scenarios and is scalable for different sources of data with varying numbers of bands. The proposed method is compatible with an embedded implementation for state-of-the-art on board hardware, a first for a ML hyperspectral data compression method. In terms of coding performance, our proposal surpasses established lossy methods such as JPEG 2000 preceded by a spectral Karhunen-Loève Transform (KLT), in clusters of 3 to 7 bands, achieving a PSNR improvement of, on average, 9 dB for AVIRIS and 3 dB for Hyperion images.