Computer Vision & ML
Neural networks, image classification, object and anomaly detection, medical imaging.
AI & Computer-Vision Engineer
From medical image processing and neural networks to embedded systems, I build intelligent systems that work in the real world.
Currently working & involved
My own research
S2P-Net is my own neural network. It recognizes objects reliably even when they are rotated and only a handful of training images exist, by combining spectral and spatial features in a polar representation.
arXiv · 2026 · First author
Vision Lab
This portrait runs through real image processing, live in your browser. Move the cursor across it or pick a mode.
A project close to my heart
An assistive app for people with learning difficulties. It guides through everyday tasks in easy language, and the camera confirms every step in the real moment. No cloud, no profit.
About
I'm Albert Heruth, I combine applied mathematics with hands-on engineering. My focus is computer vision and machine learning, from the idea all the way to systems running on real hardware.
Alongside my work on automation and camera inspection, I build my own CV models, embedded prototypes and tools, happiest where hardware meets intelligent software.
Two themes drive me most: my own research, like my S2P-Net network, and technology that helps people in everyday life, such as my assistive app Skill-Lens.
I don't just look for the bug, I look for the solution. A well-understood, well-documented system beats a quick hack.
Capabilities
Three areas I am at home in, from the first idea to a system that runs.
Neural networks, image classification, object and anomaly detection, medical imaging.
ESP32, Raspberry Pi, sensors and real-time signal processing, where software meets hardware.
Linux, Docker, networking and clean, maintainable backends & tooling.
Selected projects
A glimpse of my work. Find everything in the archive.
A minimal, human-readable DSL for describing computer-vision pipelines, with variables, loops, plugins, a linter and a browser-based visual flow editor (35 commands, 253 passing tests).
A computer-vision platform supporting instrument inspection in a medical and surgical context.
An interactive tool for debugging and visualizing CNNs with Grad-CAM, feature maps and misclassification analysis.
Interested in working together or curious about my projects? I would love to hear from you.
Get in touch