Miguel Mendez 🌊

Managing Multiple GitHub Accounts with SSH Learn how to manage multiple GitHub accounts on the same computer using SSH keys and Git configuration. This step-by-step guide covers SSH key setup, Git configuration, and automating repository acces...

miguel-mendez-ai.com/2025/02/19/c...

LLM Preference Alignment This post covers LLM preference alignment, focusing on key techniques like PPO, DPO, and ORPO. Understand how large language models (LLMs) can be aligned with human preferences. Learn about the latest...

I have been reviewing LLM alignment techniques recently and decided to summarize my thoughts in this post:
miguel-mendez-ai.com/2025/01/07/p...

We live in weird times... You can create a video from a text description, but it's almost impossible to render equations responsively on a web page

Comparing CPU vs GPU Performance in PyTorch A profiling comparison between CPU and GPU performance when normalizing images in PyTorch. There are several factors to consider when optimizing preprocessing pipelines, such as data types, data trans...

I’ve just published a new post where I explore the performance differences between CPU and GPU for image normalization in #PyTorch. Have you ever considered the importance of data types and data transfer between the CPU and GPU?
miguel-mendez-ai.com/2024/12/05/g...

Check out this BEAUTIFUL interactive blog about cameras and lenses

ciechanow.ski/cameras-and-...

Of course, manually using point correspondence is impractical for real-world use cases. We have trained models to automatically compute homography, which is essential when processing 30fps videos of 90-minute games across >90 leagues efficiently!

Image Registration in Sports Analytics: Explore computer vision techniques for mapping NFL field images to real-world coordinates. Learn how to develop a web app that uses homography to transform video feed data into precise spatial insight...

After some people asked about the difficulty of this process, I created a super simple web app to demonstrate how to manually obtain a homography using the point correspondence method. You can then use it to project the pitch lines onto the original image.
miguel-mendez-ai.com/2024/02/07/n...

Creating Better Data: AI & Homography Estimation - Hudl Statsbomb | Data Champions We use AI in our data collection process to maximise the quality of our data. Here we explain Homography Estimation and how we use it.

I wrote this short post one year ago explaining how we map player locations in sports videos using the Homography Matrix. It's super easy to read and provides a great intuition for how useful computer vision can be in sports analytics.
statsbomb.com/articles/foo...

I totally get your point about analytics, but I actually find them fun! While I don't obsess over device types or traffic sources, I enjoy watching my posts' view counts and their evolution over time. It's particularly interesting when an old post from years ago suddenly becomes popular again

No GPS required: our app can now locate underground trains Bad service in the subway? Our offline motion detection shows where you are between stations, and reminds you when your stop’s up next

Great read! It must have been a lot of fun collecting this ground truth data 🚋

blog.transitapp.com/go-undergrou...

This has been incredibly helpful to me!

Miguel Mendez | Personal Webpage Miguel Mendez personal website. Posts about Machine Learning and Deep Learning.

👋 First post here! I work as a Computer Vision Engineer at StatsBomb. I also maintain a personal blog where I occasionally share random things I find interesting:
miguel-mendez-ai.com