About

Condados is a practical resource for learning computer vision: interactive demos, the real math, and runnable code for every core idea, with honest benchmarks on edge and low-compute hardware. Understand it, run it, trust the numbers.

It's written by Luis Condados, a computer vision engineer. Seven years of it, most of them on the unglamorous side of the field: getting models out of the notebook and onto hardware that has a thermal limit, a memory budget, and someone paying for every inference.

That work has mostly happened in three places.

  • Warehouse and logistics operations, where cameras watch people and machines share a floor, and the compute is whatever box already fits in the rack.
  • Construction and infrastructure, where the output isn't a bounding box, it's a measurement someone has to sign off on.
  • Sports, where the thing that matters is a few dozen pixels across and moving faster than your frame rate.

The through-line is that the interesting problem is rarely the model. It's the quantization that costs more accuracy than the architecture ever gained. The retraining loop nobody built, so accuracy rots for six weeks before anyone notices. The classical computer vision that would have solved it with no network at all, for a fraction of the compute.

That's what this site is about, and it's why every benchmark here names the hardware it ran on and publishes the configs. You should be able to disagree with a number and prove it.

Luis is an Intel Software Innovator for OpenVINO, and holds a B.Sc. in Computer Engineering from UFRN. He's based in Natal, Brazil, on UTC−3.

Working together

Alongside the writing, I take on a small number of consulting engagements a year, mostly around getting vision systems into production and keeping them accurate once they're there. If that's the problem you have, email me at luis@condados.ai and tell me what's breaking.

Get in touch