Adrian Rosebrock, PhD

Welcome! I’m a machine learning researcher, quantitative trader, and software engineer.

I’m known online for my work on Computer Vision and Deep Learning (PyImageSearch, acquired 2021), Python and software engineering, and Quantitative Finance (WheelMetrics).

A few things about me:

1. I grew up spending my time on the Internet in the 90s in rural Maryland, USA.
2. I’m a programmer with a PhD in computer science with a focus in computer vision and machine learning.
3. During the final year of grad school I started PyImageSearch in my college apartment, ran it for eight years, and was then acquired for a life-changing exit when I was 32 years old.
4. I’ve authored 11 books, 2 courses, and 500+ tutorials on computer vision, deep learning, and OpenCV.
5. After PyImageSearch was acquired I spent years teaching myself quantitative trading and investing. I now run my own fund (AltFnd Capital) and manage the portfolio, leading the full investment and research cycle: hypothesis generation, simulation, backtesting, live deployment, and risk management.
6. Music is a huge part of my life. I grew up listening to punk, ska, and hardcore music. Now, in my late 30s, I predominately listen to EDM.
7. I speak English and 中文 (HSK2, actively learning), and have a deep obsession with 功夫茶, particularly 老白茶 from the 福鼎 and 云南 regions.

Connect

• LinkedIn
• GitHub
• Google Scholar
• Email

Computer Vision, Deep Learning, and OpenCV

• PyImageSearch (acquired): Helped millions of developers, researchers, and students learn computer vision, deep learning, and OpenCV. Authored 11 books, 2 courses, and 500+ tutorials. Acquired after eight years running it. The site is still live (operated by new owners), but I haven’t been with the company since December 2021 (circa 2014-2021).
• imutils: Python library of convenience functions to make basic image processing functions with OpenCV easier. Became quite popular despite its simplicity. Still actively used, but part of the PyImageSearch organization now (circa 2014-2021).
• bat-country: A lightweight, extendible, easy to use Python package for deep dreaming (a popular technique for visualizing what CNNs “see” back in 2014 based on the AlexNet and GoogLeNet architectures) with Caffe and CNNs. Named after Hunter S. Thompson’s Fear and Loathing in Las Vegas for obvious reasons (circa 2014-2015).
• color-transfer: Python and OpenCV package used to perform color transfer between images using the Reinhard et al. (2001) method. Very likely the first project I ever published on GitHub. Also became quite popular due to the success of PyImageSearch (circa 2014).

Quantitative Finance and Research

• WheelMetrics: Options trading research and education platform focused on teaching retail traders how to successfully apply the Wheel Strategy. Each week I publish my weekly stock screener, pre-filtered options chains, and educational content backed by my own live-traded, independently verified track record. Built on the same Python-based research platform I use to run AltFnd Capital (circa 2025-present).
• PythonFinTech (website): My first attempt at merging the two halves of my career: the PyImageSearch blog/teaching playbook applied to quantitative finance. I published Python tutorials on fintech and trading and tracked a live portfolio against the S&P 500 with full risk metrics (Sharpe, Sortino, max drawdown, etc.). I shut the site down as I came to the realization that “code explainer blogs” really aren’t relevant in the age of LLMs. Still, the lessons learned here directly shaped WheelMetrics (circa 2024-2025).
• PythonFinTech (code): Jupyter Notebooks and code associated with the tutorials on PythonFinTech. The code primarily focuses on fetching and analyzing market data with Pandas, computing risk and performance metrics, and screening stocks for technical setups. Basically, teaching the fundamentals of quantitative finance with an engineer-first mindset (circa 2024-2025).

Older Projects

Older projects that may have had some impact, or I just find interesting and worth highlighting.

• rescue-ripple: Part of my research on Zero-Shot, Prompt-Based Classification with LLMs for Disaster Tweet Analysis. Soon after OpenAI released their APIs, I decided to evaluate both zero-shot and embedding classification on a real-world dataset (HumAID Twitter dataset of natural disaster tweets). This work demonstrates that zero-shot and embedding methods are comparable (and, in some cases, outperform) traditional hand-crafted tf-idf + SVM feature pipelines, and require far less effort than their traditional counterparts (circa 2023).
• ID My Pill (2023): For fun, I returned to my original ID My Pill project over a decade later and was able to successfully implement the project thanks to advances in computer vision and artificial intelligence. This version can identify prescription pills from both a photo and plain-language description. Utilizes a novel discriminative co-occurrence search and ranking algorithm across pill shape, imprint, and color. Works on 20,000+ US pills. Implemented as a Python library + Django app + Custom GPT interface (circa 2023).
• DateTimeCoding: After PyImageSearch was acquired I was both exhausted and bored, which is apparently the mental state required to voluntarily start a blog about datetime programming (i.e., timezones, DST, leap seconds, the whole miserable landscape). The project lasted all of six months, which honestly might be longer than the subject deserved (circa 2022).
• ID My Pill (2013): My first attempt at identifying prescription pills in images. Built during early grad school using classical computer vision techniques of the era: hand-crafted shape, color, and texture features + descriptors. It worked well enough to be interesting (and make some money), but more importantly, it taught me the limits of computer vision during the pre-deep learning era (circa 2013).
• Chic Engine: Chic Engine was a visual fashion search engine. You provided it with a photo of a clothing item you wanted to buy, the search engine found identical/visually similar items for sale, and then presented the search results to you. Built during my first few semesters in grad school. Never made me rich, but it covered my car payments most months, which, for a college student is a pretty big deal (circa 2012-2013).
• RateMyTeachers: The archived source for a reusable PHP/CodeIgniter rating engine that powered RateMyTeachers.com. As a single developer, I rewrote the entire platform at 19-20 years old to reliably serve 100,000+ unique visitors per day, replacing a legacy codebase that was going down 2-3 times a week. Featured hand-tuned SQL, offline batch processing for the heavy computations, and an automated content-moderation system built on n-gram text classification with Java and Weka (circa 2011-2013).