Miguel Méndez

Bio

I am a Machine Learning Engineer at Statsbomb. I love creating systems based on data that can interact with the real world. My research interests lie in the general area of machine learning, particularly in deep learning and its applications in object detection, image generation and instance segmentation.

Prior to coming to Statsbomb, I worked at Gradiant as a Machine Learning Research Engineer and at Desigual Headquarters as a Data Science consultant. I spent one year in Barcelona and one year in the United States (Purdue University) to do a M.Sc. in Artificial Intelligence. I got my B.Sc. from University of A Coruña after an exchange year in Norway.

News

• September 2023. Published a blog post about homography estimation in Statsbomb blog [English, Spanish].

• February 2022. I am joining Statsbomb as a Machine Learning Engineer .

• April 2021. Drone vs Bird paper is out! Check our awarded solution.

• February 2021. We have open sourced Pyodi our Python Object Detecion Insights library.

• September 2020. We have won Drone vs Bird Detection Challenge. More info here.

• September 2019. Paper from IEEE AVSS2019 Drone-vs-Bird Detection Challenge is out.

Posts

• 

  Deploying MLFlow in AWS with Terraform

  Streamline Your ML Workflows: A Simple Guide to MLFlow Deployment on AWS

• 

  Multi-Object Tracking by Detection: A Comprehensive Guide

  Exploring the Advancements and Accuracy in Multi-Object Tracking Technologies

• 

  Creating Better Data: AI & Homography Estimation

  The cornerstone of Statsbomb tracking data collection methodology

• 

  Torchserve in Computer Vision: REST vs gRPC

  Benchmarking protocols performance for sending images

• 

  CNNs & Transformers Explainability: What do they see?

  A simple explanation and visualization of ViT attentions and ResNet activations

• 

  Using Ansible and Python to monitor my servers

  Code your own scripts in Python, deploy them using Ansible & Conda and get real time reports from your server

See more posts
• 

  DVC + MMdetection

  A guide to train, monitor, compare and evaluate your pytorch object detection models

• 

  Export to Github Cache with Docker Buildx

  We can finally use Docker buildx cache-to gha with build-push action and it is blazingly fast!

• 

  Version control your dataset with DVC

  Use DVC and Git for tracking changes on your machine learning dataset

• 

  Reduce Actions time with Docker and Github Cache

  Use Github Actions cache and Docker to reduce time installing Conda dependencies

• 

  Generating FIFA 19 players with VAEs and Tensorflow

  Learn to generate your own footbal players

• 

  VAEs! Generating images with Tensorflow

  Learn how to code a Variational Autoencoder

• 

  The theory behind Variational Autoencoders

  Learn the math

Vitæ

• Vigo, ESP (Remote)

  Feb'22 - now
  

  Machine Learning Engineer - Statsbomb

• Vigo, ESP

  Apr'19 - Jan'22
  

  Machine Learning Research Engineer - Gradiant

  Research and implementation of Deep Learning based models for object detection and semantic segmentation. Optimization of these solutions for being embedded on UAVs (Nvidia Jetson) or deployed in the cloud. Leadership, coordination and development of the teams AI methodologies and infrastructure.

• Barcelona, ESP

  Feb'18 - Apr'19
  

  Data Scientist, Consultant - Desigual, InnoIT

  Developed machine learning models for analyzing stock distribution and sales / returns forecasting. Research on using CNN for database image retrieval and user recommendations.

• Indiana, USA

  Aug'16 - Jun'17
  

  Master in Artificial Intelligence (Study Abroad) - Purdue University

  I was a member of Purdue Datalab and collaborated with PhD students by presenting, exploring and discussing new topics on AI research. Also took data mining, machine learning courses and finish my Master Thesis: Feature construction and classification on Time Series.

• Barcelona, ESP

  Sep'15 - Jun'17
  

  Master in Artificial Intelligence - Polytechnic University of Catalonia

  This was the moment when my adventure with AI began. I remember being totally impressed with the CNNs of the moment! I took courses that were mainly related with supervised / unsupervised learning approaches.

• Stavanger, NOR

  Aug'14 - Jun'14
  

  B.Sc. Student (Study Abroad) - University of Stavanger

  I got awarded with Erasmus and NILS scholarship. I did my final project with three ingredients: Optical Flow, a GoPro camera and my own bike.

• A Coruña, ESP

  Sep'11 - Jun'16
  

  B.Sc. Student - University of A Coruña

  This is where I learned about data structures, algorithm complexity and so many other things that I use so often that still surprise myself from time to time.

Publications

• Drone vs. Bird Detection: Deep Learning Algorithms and Results from a Grand Challenge

  Sensors: Special Issue Deep Learning Based UAV Detection, Classification, and Tracking

  Coluccia, Angelo and Fascista, Alessio and Schumann, Arne and Sommer, Lars and Dimou, Anastasios and Zarpalas, Dimitrios and Méndez, Miguel and de la Iglesia, David and González, Iago and Mercier, Jean-Philippe and Gagné, Guillaume and Mitra, Arka and Rajashekar, Shobha

• Drone-vs-Bird Detection Challenge at IEEE AVSS2019

  2019 16th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS)

  A. Coluccia and A. Fascista and A. Schumann and L. Sommer and M. Ghenescu and T. Piatrik and G. De Cubber and M. Nalamati and A. Kapoor and M. Saqib and N. Sharma and M. Blumenstein and V. Magoulianitis and D. Ataloglou and A. Dimou and D. Zarpalas and P. Daras and C. Craye and S. Ardjoune and D. De la Iglesia and M. Méndez and R. Dosil and I. González

Other Projects

• Pyodi: Python Object Detection Insights

  A simple tool for explore your object detection dataset. The goal of this library is to provide simple and intuitive visualizations from your dataset and automatically find the best parameters for generating a specific grid of anchors that can fit you data characteristics

  Tags: #deep-learning #computer-vision #object-detection #pytorch

• Feature construction and classification on Time Series

  Collaborated with Federal Aviation Administration (FAA) in PEGASAS project with the aim of enhance general aviation safety. Developed a model which could classify flights in different categories (using data obtained from airborne sensors) through an unsupervised learning stage and an active learning procedure. Tools: Python, Scikit-Learn, Pandas, C, OpenMPI, Java
  Presented in PEGASAS 2017 meeting at Texas A&M University

  Tags: #machine-learning #scikit-learn

• Instrument recognition using Neural Networks

  System for musical instrument recognition using NNs with over 10000 audio samples based on Fourier Transform input features. Created our own NN library in Python from scratch.

  Tags: #python #numpy #fftransform

• Corruption simulation in Social Networks (Twitter)

  Analysis using Twitter public API where we focused in understanding how politicians relate in the social network. We performed a Monte Carlo simulation of corruption behavior spreading with the aim of observing how corruption flows starting from some seed users that were imprisoned or involved in some case.

  Tags: #twitter #python #numpy

• Flight Safe Android App

  Developed Android app for collecting AHRS data during flights to facilitate the creation of a dataset for future experiments. This app was used by pilots and students that flew from Purdue’s airport.

  Tags: #java #android #sensors