The study blog from marearts.com
Computer Vision & Machine Learning Research Laboratory
marearts-tinydeform is our first open-source CUDA kernel. It runs the value projection and multi-scale deformable attention of a DETR-style decoder (RT-DETR, D-FINE, DINO, Deformable-DETR) as one call, gives the same result as the usual PyTorch code, and is up to 4.7× faster than the original Deformable-DETR CUDA op when there are few queries. Apache-2.0, free for any use.
A mock is a stand-in you put where a real function or object used to be, for as long as a test
runs. It answers the way you tell it to, it remembers every call it got, and when the test ends the
original comes back. This guide covers both halves — controlling and recording — and then the
part most tutorials skip: wraps=, a spy that lets the real code run.
def sensor_id(sensor: Sensor | int | None) -> int has three bars in one signature and
looks like bit arithmetic. It is not arithmetic and it is not new syntax to memorise: each bar is the
word or, and the whole annotation is one sentence — "a Sensor, or an
int, or nothing".