Why a One-Pixel Shift Can Change a CNN Prediction
Why strided downsampling can make CNN predictions sensitive to a one-pixel shift—and what anti-aliasing can and cannot fix.
8 hours ago2 min read
Why Changing Batch Size Can Change a Prediction
Why batching can change floating-point results—and how tiny score differences can become different inspection decisions.
Sep 282 min read
What Does “Close” Mean? Mahalanobis Distance for Visual Defects
Why equal Euclidean distances can hide different kinds of anomalies—and how covariance, whitening, and Mahalanobis distance change visual inspection.
Sep 142 min read
Non-Blocking Isn’t Parallel: A PyTorch Detail That Matters
Why non_blocking=True removes a CPU-side wait but does not guarantee GPU overlap—and how pinned memory, CUDA streams, and batching make the difference.
Sep 101 min read
One Model, Many Cameras: Why Dynamic Resolution Matters in VLMs
How dynamic resolution helps VLMs handle images from different cameras—and why patch counts, preprocessing, and compute budgets still matter.
Sep 72 min read
Beyond AUROC: Evaluating Visual Anomaly Detection for Real-World Production
High benchmark performance does not guarantee a reliable inspection system. This article presents a production-oriented framework for evaluating visual anomaly detection across thresholds, distribution shift, error costs, latency, validation, and monitoring.
Sep 33 min read
How I Completed Udacity's Agentic AI Nanodegree in 6 Hours
It was a Sunday morning in Munich. I had a cup of double espresso in hand, my laptop open, and a simple question in my head: how fast can I actually finish this? Six hours later, I had my answer — and a Udacity Nanodegree certificate to prove it. The Backstory I'll be honest. I didn't set out to break any records. I'm the CTO and Co-founder of 36ZERO Vision, a Munich-based company that builds AI-powered quality inspection systems for manufacturers like Siemens, Bosch, and Mag
Feb 14 min read
