Become a sponsor to CondadosAI

Computer vision, measured rather than asserted

CondadosAI publishes open-source computer vision work together with the numbers behind it —
the code, the benchmark configs, and the hardware each result was measured on. Detection,
segmentation, tracking, structure from motion, local features: the write-ups at
condados.ai ship with a repository you can rerun.

The flagship is modern-yolonas — YOLO-NAS
rebuilt as plain PyTorch, with no factories, registries, or config frameworks, loading the
original pretrained weights unchanged.

Sponsorship pays for GPU time. The pretrained YOLO-NAS COCO checkpoints everyone
uses are Deci's, and their license forbids commercial use. The only way out is to train COCO
weights from scratch and publish them under Apache-2.0 — a compute bill, not a coding problem:
about $80–120 of spot GPU time for YOLO-NAS-S, several times that for L. Publishing honest
latency and accuracy numbers is rented time too.

Everything funded here is published for everyone: the weights, the numbers, the failures, and
the configs. There is no sponsors-only tier of results.

💬 Looking for help with your own models instead? That is not sponsorship — book a call: cal.com/luis-condados. 🧾 Companies that need an invoice or a specific benchmark run: open an issue and we will sort it out there.