News :

|

Miguel Crespo

Hi! My name is Miguel. I studied a Bachelor's degree in Computer Science and a Master's degree in Biomedical Engineering, both at Universidad de Zaragoza (Spain). During that time and after, I have been a junior researcher at the Graphics & Imaging Lab under the supervision of Prof. Adolfo Muñoz and Prof. Adrian Jarabo.

My work revolves around the different aspects of physically-based light transport, including simulation and reconstruction, with mention on both transient and computational imaging. Additionally I am interested in the intersection of machine learning with the previous fields. Nevertheless, I’m always looking forward to learning something new!

  • Twitter
  • Linkedin
  • GitHub

Publications

Primary-Space Adaptive Control Variates using Piecewise-Polynomial Approximations

Miguel Crespo , Adrian Jarabo , Adolfo Muñoz
Arxiv : 2008.06722 , 2020
Arxiv
  • PDF
  • Project page

Vision through Turbid Media using Phasor Fields

Miguel Crespo , Adrian Jarabo , Diego Gutierrez
Zaguan - Universidad de Zaragoza , 2019
MSc Thesis
  • PDF
  • Honorary Mention | (Spanish only)

Adaptive Rendering using Quadrature Techniques

Miguel Crespo , Adolfo Muñoz
Zaguan - Universidad de Zaragoza , 2018
BSc Thesis
  • PDF
  • (Spanish only)

Projects

    Blender Exporter for Mitsuba 1

    Fully fledged Blender exporter with extended support of cycle materials, participating media and editor of curves to export motion blur for both geometry and camera. Also, added support for true Point Lights.

    • Code
  • ANZEIGE: Interactive Results Suite

    Automatic generator of a webpage with comparisons of renders with support of visualizer (including insets), runtime-generated several error metrics, video player with slider and error metric table.

    • MOVBOT: Zaragoza Public Transport Telegram Assistant

      Telegram bot that allows to direct query, fuzzy search or GPS-locate stops of both bus and tram in Zaragoza. Allows to track easily times than the original webpage of the city council.

      • J.A.R.V.I.S: Discord Bot based in NLP and Voice Control

        Fully-fledge Discord bot with support for answer queries both writen and speaked that can run in low-tier cloud systems. Added voice control is a marvellous when you are ingame and need to manage something from the server (or change the playlist).

        • LIGHTNING: Physically-based Renderer on CPU

          Fully CPU-parallel Physically-based renderer featuring both unidireccional Path Tracing and Photon Mapping. As additional features, next event estimation, acceleration structures, participating media and several BRDF. This project was awarded with a honor mention and was the winner of the best looking render competition.

          • CRISTOBAL: Mindstorm-based fully fledge robot OS

            The system features dynamic path planning and awareness of the surrounding world by combining odometry of the wheels, gyroscope, compass and sonar. To reduce shift in data, it features a sensor-fusion approach using Extended Kalman Filter, allowing to increase the accuraty of the position. This project was awarded with the mention of honor as the best approach.

            • LITTLE-LENG: Full AST Compiler for Custom Language

              Full compiler using JavaCC with a custom-made abastract syntax tree. Compiler features constant-time aggregation of operations, strong typing, usefull error/warning messages (including suggestions of misspell names), removal of unused parameters, optimization of math operators and emojis.

              • VERWEISE: Next-gen Bibliography Manager

                Cross-platform bibliography manager with a focus in usability and management of larger libraries. Future features will support sharing between users and tentative ML approach to discover and extract information.

                • BASKET-VR: Assignment for VR Course

                  Assignment for the VR Course where the students learn about movement in VR enviorements and how to interact with the world using controllers. To make it fun, students end playing virtual basketball.

                  • Fourier Classifier based on CNN Arquitecture

                    Improved solution that changes the net architecture to use Convolutions, allowing the system to work with crops of the signal instead of the full data. This solution was originated as an improvement over previous solution to allow its use in realtime situations (as games livestreams).

                    • UCODE 2019 - Fourier Classifier ML

                      Our project submitted to the hackathon competition where we solved the Adidas Advanced Division's problem about classifying movement sequences using Machine Learning applied in frequency space. We achieved a 94% accuraty.

                      • UCODE 2018 - Indoors Beacon-based Location

                        Our project submitted to the hackathon competition where we designed and implemented an indoors solution for tracking users when GPS is not an option. Our solution works with Bluetooth technology and allows to guide users given their positions and preferences.

                        About me

                        Research Experience

                        • Graphics & Imaging Lab - Junior Researcher2018 - 2020

                          Universidad de Zaragoza (Spain)

                        Education

                        • MSc - Biomedical Engineering 2018 - 2019

                          Universidad de Zaragoza (Spain)

                        • BSc - Computer Science 2014 - 2018

                          Universidad de Zaragoza (Spain)

                        Contact

                        Dpto. Informática e Ing. de Sistemas. Lab. 2.06.
                        C/ María de Luna, 1. Ed. Ada Byron.
                        50018, Zaragoza
                        SPAIN
                        35 32 67 679 (43)
                        se.razinu@opsercm

                        Gallery

                        photophoto
                        photophoto
                        photophoto
                        photophoto
                        photophoto
                        photophoto
                        photophoto
                        photophoto
                        photophoto
                        photophoto
                        photophoto

                        © Miguel Crespo 2021 - Adapted from Dimension. Design: HTML5 UP.