Agreed, this article is pure misrepresentation. However, are DETRs performing better than YOLO? That's the salient question, for me.

DETR model have outperformed YOLO models for a while, but they have been much slower making them impractical for real time detection.

A small percent difference means nothing at all.

Your camera will be very different from the images the models have been trained on, so you'll have to do fine tuning. So the only thing that matters is this phase.

Plus, there are models that run device, on servers etc.