Pasar al contenido principal

GENERACION PROCEDURAL DE ESCENARIOS AUMENTADOS CON ANOTACION DE DATOS AUTOMATICA

PID2020-115734RB-C22

•
Nombre agencia financiadora Agencia Estatal de Investigación
Acrónimo agencia financiadora AEI
Programa Programa Estatal de I+D+i Orientado a los Retos de la Sociedad
Convocatoria Proyectos de I+D+i (Generación de Conocimiento y Retos Investigación)
Año convocatoria 2020
Plan Estatal 2017-2020 Plan Estatal de Investigación Científica y Técnica y de Innovación
Centro beneficiario UNIVERSIDADE DA CORUÑA

Resultados relacionados

Resultados totales (Incluyendo duplicados): 6
Encontrada(s) 1 página(s)

RUC. Repositorio da Universidade da Coruña
Publicación Artículo científico (JournalArticle)

Neural James-Stein Combiner for Unbiased and Biased Renderings

  • Gu, Jeongmin
  • Iglesias-Guitian, Jose A.
  • Moon, Bochang
This is the author's version of the work. It is posted here for your personal use. Not for redistribution. The definitive Version of Record was published in https://doi.org/10.1145/3550454.3555496.", [Abstract]: Unbiased rendering algorithms such as path tracing produce accurate images given a huge number of samples, but in practice, the techniques often leave visually distracting artifacts (i.e., noise) in their rendered images due to a limited time budget. A favored approach for mitigating the noise problem is applying learning-based denoisers to unbiased but noisy rendered images and suppressing the noise while preserving image details. However, such denoising techniques typically introduce a systematic error, i.e., the denoising bias, which does not decline as rapidly when increasing the sample size, unlike the other type of error, i.e., variance. It can technically lead to slow numerical convergence of the denoising techniques. We propose a new combination framework built upon the James-Stein (JS) estimator, which merges a pair of unbiased and biased rendering images, e.g., a path-traced image and its denoised result. Unlike existing post-correction techniques for image denoising, our framework helps an input denoiser have lower errors than its unbiased input without relying on accurate estimation of per-pixel denoising errors. We demonstrate that our framework based on the well-established JS theories allows us to improve the error reduction rates of state-of-the-art learning-based denoisers more robustly than recent post-denoisers, We appreciate the anonymous reviewers for the constructive comments. We also thank the following authors and artists for each scene: Mareck, SlykDragko, Wig42, NovaZeeke and thecali (training scenes in Fig. 9), aXel (Glass-of-water), Cem Yuksel (Curly-hair), NewSee20135 (Staircase), Ondřej Karlík (Pool), Tiziano Portenier (Bookshelf for the Mitsuba porting), and Christian Schüller (Dragon). Bochang Moon is the corresponding author of the paper. This work was supported by the National Research Foundation of Korea (NRF) funded by the Korea government (MSIT) (No. 2020R1A2C4002425) and Ministry of Culture, Sports and Tourism and Korea Creative Content Agency (No. R2021080001). Jose A. Iglesias-Guitian was supported by a 2021 Leonardo Grant for Researchers and Cultural Creators, BBVA Foundation. He also acknowledges the UDC-Inditex InTalent programme, the Spanish Ministry of Science and Innovation (AEI/PID2020-115734RB-C22 and AEI/ RYC2018-025385-I), Xunta de Galicia (ED431F 2021/11) and EUFEDER Galicia (ED431G 2019/01)., Korea. Korea government (MSIT); 2020R1A2C4002425, Korea. Ministry of Culture, Sports and Tourism; R2021080001, Xunta de Galicia; ED431F 2021/11, Xunta de Galicia; ED431G 2019/01
RUC. Repositorio da Universidade da Coruña
Publicación Artículo científico (JournalArticle)

All for one, and one for all: UrbanSyn Dataset, the third Musketeer of synthetic driving scenes

  • Gómez Zurita, José Luis
  • Silva, Manuel
  • Seoane, Antonio
  • Borràs, Agnés
  • Noriega, M. A.
  • Ros, German
  • Iglesias-Guitian, Jose A.
  • López, Antonio M.
This manuscript version is made available under the CC-BY-NC-ND 4.0 license https://creativecommons.org/licenses/by-nc-nd/4.0/. This version of the article has been accepted for publication in Neurocomputing (ISSN 1872-8286)., [Abstract]: We introduce UrbanSyn, a photorealistic dataset acquired through semi-procedurally generated synthetic urban driving scenarios. Developed using high-quality geometry and materials, UrbanSyn provides pixel-level ground truth, including depth, semantic segmentation, and instance segmentation with object bounding boxes and occlusion degree. It complements GTAV and Synscapes datasets to form what we coin as the 'Three Musketeers'. We demonstrate the value of the Three Musketeers in unsupervised domain adaptation for image semantic segmentation. Results on real-world datasets, Cityscapes, Mapillary Vistas, and BDD100K, establish new benchmarks, largely attributed to UrbanSyn. We make UrbanSyn openly and freely accessible, This work has been supported by the Spanish grants Ref. PID2020-115734RB-C21 (ADA/SSL-ADA subproject) and PID2020-115734RB-C22 (ADA/PGAS-ADA subproject), both funded by MCIN/AEI/ 10.13039/501100011033. Antonio M. López acknowledges the financial support to his general research activities given by ICREA under the ICREA Academia Program, Spain. CVC’s authors acknowledge the support of the Generalitat de Catalunya CERCA Program and its ACCIO agency to CVC’s general activities. Jose A. Iglesias-Guitian acknowledges the financial support to his general research activities given by UDC-Inditex InTalent programme, the Spanish Ministry of Science and Innovation (AEI/RYC2018-025385-I), and Xunta de Galicia, Spain (ED431F 2021/11, EU-FEDER ED431G 2019/01), Xunta de Galicia; ED431F 2021/11, Xunta de Galicia; ED431G 2019/01
RUC. Repositorio da Universidade da Coruña
Publicación Contribución de congreso (ConferenceOutput)

Highway Remodeling: Harnessing Georeferenced Data for Procedural Modeling

  • Noriega, M. A.
  • Ribelles, José
  • Seoane, Antonio
  • Martínez, Margarita
  • Taibo, Javier
  • Iglesias-Guitian, Jose A.
Cursos e Congresos, C-155, [Abstract] : This paper introduces a novel procedural modeling system for generating 3D highway models, leveraging real-world data inputs such as highway layouts and manual annotations of essential elements. Our highly parameterized system facilitates easy customization, including modifications to the number of lanes and other key features, broadening its applicability across diverse domains, from urban planning and transportation engineering to virtual simulations, Generalitat Valenciana; CIBEST/2022/139, Xunta de Galicia; ED431F 2021/11, This work has been supported by the Spanish Ministry of Science and Innovation (AEI/PID2020-115734RB-C22). The whole team also wants to acknowledge the support provided by Side Effects Software Inc. in developing our work. Jose Ribelles was supported by Generalitat Valenciana (CIBEST/2022/139). Javier Taibo and J.A. Iglesias-Guitian were supported by Xunta de Galicia (ED431F 2021/11). Additionally, J.A. Iglesias-Guitian also acknowledges the UDC-Inditex InTalent programme and the Ministry of Science and Innovation (AEI/RYC2018-025385-I)
RUC. Repositorio da Universidade da Coruña
Publicación Artículo científico (JournalArticle)

Exploring the effects of synthetic data generation: a case study on autonomous driving for semantic segmentation

  • Silva, Manuel
  • Seoane, Antonio
  • Mures, Omar A.
  • López, Antonio M.
  • Iglesias-Guitian, Jose A.
Financiado para publicación en acceso aberto: Universidade da Coruña/CISUG, [Abstract]: Rendering 3D virtual scenarios has become a popular alternative for generating per-pixel-labeled image datasets, especially in fields like autonomous driving. The approach is valuable for training neural perception models, such as semantic segmentation models, particularly when data might be scarce, expensive, or difficult to collect. However, fundamental questions persist within the research community regarding the generation and processing of these synthetic images, particularly a better understanding of the key factors influencing the performance of deep learning models trained with such synthetic images. In response, we conducted a series of experiments to elucidate the impact that common aspects involved in the generation of rendered synthetic images may have on the performance of neural semantic segmentation tasks. Our study used a recent autonomous driving synthetic dataset as our main testbed, allowing us to investigate the effect of different approaches when modeling their geometric, material, and lighting details. We also studied the impact of rendering noise, typically produced by path-tracing algorithms, as well as the impact of using different color transformations and tonemapping algorithms., Open Access funding provided thanks to the CRUE-CSIC agreement with Springer Nature. This work has been supported by Spanish grants Ref. PID2020-115734RB-C22 and PID2020-115734RB-C21, both funded by MCIN/AEI/10.13039/501100011033. Manuel Silva acknowledges Xunta de Galicia (Consellería de Cultura, Educacion e Universidade) for his predoctoral grant (ED481A-2023-191). Antonio M. López acknowledges the financial support to his general research activities given by ICREA under the ICREA Academia Program and the support of the Generalitat de Catalunya CERCA Program and its ACCIO agency to CVC’s general activities. J.A. Iglesias-Guitian also acknowledges the UDC-Inditex InTalent program, the Ministry of Science and Innovation (AEI/ RYC2018-025385-I) and Xunta de Galicia (ED431F 2021/11)., Xunta de Galicia; ED481A-2023-191, Xunta de Galicia; ED431F 2021/11
RUC. Repositorio da Universidade da Coruña
Publicación Artículo científico (JournalArticle)

Should I render or should AI Generate? Crafting Synthetic Semantic Segmentation Datasets with Controlled Generation

  • Mures, Omar A.
  • Silva, Manuel
  • Lijó-Sanchez, Manuel
  • Padrón, Emilio J.
  • Iglesias-Guitian, Jose A.
Versión aceptada, publicada en modo "Early Publication" por el editor., [Abstract]: This work explores the integration of generative AI models for automatically generating synthetic image-labeled data. Our approach leverages controllable Diffusion Models to generate synthetic variations of semantically labeled images. Synthetic datasets for semantic segmentation struggle to represent real-world subtleties, such as different weather conditions or fine details, typically relying on costly simulations and rendering. However, Diffusion Models can generate diverse images using input text prompts and guidance images, like semantic masks. Our work introduces and tests a novel methodology for generating labeled synthetic images, with an initial focus on semantic segmentation, a demanding computer vision task. We showcase our approach in two distinct image segmentation domains, outperforming traditional computer graphics simulations in efficiently creating diverse datasets and training downstream models. We leverage generative models for crafting synthetically labeled images, posing the question: “Should I render or should AI generate?”. Our results endorse a paradigm shift towards controlled generation models, Xunta de Galicia: ED431F 2021/11, Xunta de Galicia: ED431G 2019/01, The authors would like to thank José R. López from Deep Design Systems S.L. for the car parts dataset (PART). This work has been supported by Xunta de Galicia: ED431F 2021/11 and ED431G 2019/01. It was also partially supported through the research projects AEI/PID2020-115734RB-C22 and PID2022-136435NB-I00, funded by MCIN/AEI/10.13039/501100011033 and by ‘ERDF A way of making Europe’, EU. Jose A. Iglesias-Guitian also acknowledges the UDC-Inditex InTalent programme and the Spanish Ministry of Science and Innovation (AEI/RYC2018-025385-I)
RUC. Repositorio da Universidade da Coruña
Publicación Contribución de congreso (ConferenceOutput)

Understanding the Influence of Rendering Parameters in Synthetic Datasets for Neural Semantic Segmentation Tasks

  • Silva, Manuel
  • Mures, Omar A.
  • Seoane, Antonio
  • Iglesias-Guitian, Jose A.
Cursos e Congresos , C-155, [Abstract] Deep neural networks are well known for demanding large amounts of training data, motivating the appearance of multiple synthetic datasets covering multiple domains. However, synthetic datasets have not yet outperformed real data for autonomous driving applications, particularly for semantic segmentation tasks. Thus, a deeper comprehension about how the parameters involved in synthetic data generation could help in creating better synthetic datasets. This work provides a summary review of prior research covering how image noise, camera noise and rendering photorealism could affect learning tasks. Furthermore, we presents novel experiments aimed at advancing our understanding around generating synthetic data for autonomous driving neural networks aimed at semantic segmentation, Xunta de Galicia; ED431F 2021/11, This work has been supported by the Spanish Ministry of Science and Innovation (AEI/PID2020-115734RB-C22). We also want to acknowledge Side Effects Software Inc. for their support to this work. J.A. Iglesias-Guitian also acknowledges the UDC-Inditex InTalent programme, the Ministry of Science and Innovation (AEI/RYC2018-025385-I) and Xunta de Galicia (ED431F 2021/11). CITIC is funded by the Xunta de Galicia through the collaboration agreement between the Consellería de Cultura, Educación, Formación Profesional e Universidades and the Galician universities for the reinforcement of the research centres of the Galician University System (CIGUS)