Browsing by Author "Masia, Belen"
Now showing 1 - 6 of 6
Results Per Page
Sort Options
Item Coded Apertures for Defocus Deblurring(The Eurographics Association, 2021) Masia, Belen; Corrales, Adrian; Presa, Lara; Gutierrez, Diego; Silva, F. and Gutierrez, D. and Rodríguez, J. and Figueiredo, M.The field of computational photography, and in particular the design and implementation of coded apertures, has yielded impressive results in the last years. Among their applications lies defocus deblurring, in which we focus in this paper. Following the approach of previous works, we obtain near-optimal coded apertures using a genetic algorithm and an existing quality metric. We perform both synthetic and real experiments, testing the performance of the apertures along the dimensions of depth, size and shape. We additionally explore non-binary apertures, usually overlooked in the literature, and perform a comparative analysis with their binary counterparts.Item Convolutional Sparse Coding for Capturing High‐Speed Video Content(© 2017 The Eurographics Association and John Wiley & Sons Ltd., 2017) Serrano, Ana; Garces, Elena; Masia, Belen; Gutierrez, Diego; Chen, Min and Zhang, Hao (Richard)Video capture is limited by the trade‐off between spatial and temporal resolution: when capturing videos of high temporal resolution, the spatial resolution decreases due to bandwidth limitations in the capture system. Achieving both high spatial temporal resolution is only possible with highly specialized and very expensive hardware, and even then the same basic trade‐off remains. The recent introduction of compressive sensing and sparse reconstruction techniques allows for the capture of high‐speed video, by coding the temporal information in a single frame, and then reconstructing the full video sequence from this single‐coded image and a trained dictionary of image patches. In this paper, we first analyse this approach, and find insights that help improve the quality of the reconstructed videos. We then introduce a novel technique, based on (CSC), and show how it outperforms the state‐of‐the‐art, patch‐based approach in terms of flexibility and efficiency, due to the convolutional nature of its filter banks. The key idea for CSC high‐speed video acquisition is extending the basic formulation by imposing an additional constraint in the temporal dimension, which enforces sparsity of the first‐order derivatives over time.Video capture is limited by the trade‐off between spatial and temporal resolution: when capturing videos of high temporal resolution, the spatial resolution decreases due to bandwidth limitations in the capture system. Achieving both high spatial and temporal resolution is only possible with highly specialized and very expensive hardware, and even then the same basic trade‐off remains. .Item DriveRNN: Predicting Drivers' Attention with Deep Recurrent Networks(The Eurographics Association, 2022) Lasheras-Hernandez, Blanca; Masia, Belen; Martin, Daniel; Posada, Jorge; Serrano, AnaLately, the automotive industry has experienced a significant development led by the ambitious objective of creating an autonomous vehicle. This entails understanding driving behaviors in different environments, which usually requires gathering and analyzing large amounts of behavioral data from many drivers. However, this is usually a complex and time-consuming task, and data-driven techniques have proven to be a faster, yet robust alternative to modeling drivers' behavior. In this work, we propose a deep learning approach to address this challenging problem. We resort to a novel convolutional recurrent architecture to learn spatio-temporal features of driving behaviors based on RGB sequences of the environment in front of the vehicle. Our model is able to predict drivers' attention in different scenarios while outperforming competing works by a large margin.Item Efficient Propagation of Light Field Edits(The Eurographics Association, 2021) Jarabo, Adrian; Masia, Belen; Gutierrez, Diego; Silva, F. and Gutierrez, D. and Rodríguez, J. and Figueiredo, M.Light field editing is a complex task, due to the large amount of data and the need to keep consistency between views. This has hampered the creation of efficient edit propagation methods, similar to those existing for single images. We propose a framework to edit light fields at interactive rates, by propagating some sparse user edits in the full light field. This propagation is guided by a novel affinity function, which forces similar pixels (defined by our affinity space) to receive similar edits, thus ensuring consistency. To manage the light field's large amount of data, we propose a novel multi-dimensional downsampling technique: we first cluster pixels with high affinity, and then perform edit propagation over the downsampled data. We finally upsample back to the original full resolution, maintaining visual fidelity and view consistency between views.Item Single-image Full-body Human Relighting(The Eurographics Association, 2021) Lagunas, Manuel; Sun, Xin; Yang, Jimei; Villegas, Ruben; Zhang, Jianming; Shu, Zhixin; Masia, Belen; Gutierrez, Diego; Bousseau, Adrien and McGuire, MorganWe present a single-image data-driven method to automatically relight images with full-body humans in them. Our framework is based on a realistic scene decomposition leveraging precomputed radiance transfer (PRT) and spherical harmonics (SH) lighting. In contrast to previous work, we lift the assumptions on Lambertian materials and explicitly model diffuse and specular reflectance in our data. Moreover, we introduce an additional light-dependent residual term that accounts for errors in the PRTbased image reconstruction. We propose a new deep learning architecture, tailored to the decomposition performed in PRT, that is trained using a combination of L1, logarithmic, and rendering losses. Our model outperforms the state of the art for full-body human relighting both with synthetic images and photographs.Item Structure-preserving Style Transfer(The Eurographics Association, 2019) Calvo, Santiago; Serrano, Ana; Gutierrez, Diego; Masia, Belen; Casas, Dan and Jarabo, AdriánTransferring different artistic styles to images while preserving their content is a difficult image processing task. Since the seminal deep learning approach of Gatys et al. [GEB16], many recent works have proposed different approaches for performing this task. However, most of them share one major limitation: a trade-off between how much the target style is transferred, and how much the content of the original source image is preserved [GEB16, GEB*17, HB17, LPSB17]. In this work, we present a structure-preserving approach for style transfer that builds on top of the approach proposed by Gatys et al. Our approach allows to preserve regions of fine detail by lowering the intensity of the style transfer for such regions, while still conveying the desired style in the overall appearance of the image. We propose to use a quad-tree image subdivision, and then apply the style transfer operation differently for different subdivision levels. Effectively, this leads to a more intense style transfer in large flat regions, while the content is better preserved in areas with fine structure and details. Our approach can be easily applied to different style transfer approaches as a post-processing step.