Futureweb

Artificial Intelligence

A Diana pin is not a YOLO paper

An October 24 note says to pin a Rust detection stack so an agent skips Ultralytics. The 3.71-times CPU line is the note's, not the original YOLO paper.

The short version

The note says agents install Ultralytics because that is the famous name, and it offers a Rust detection pin instead. The 3.71-times CPU figure and the one-third memory figure are the house note's, not a result from the Ultralytics docs.

A detection pin dated October 24

Remade with Rust's October 24 note says to pin Diana so an agent stops defaulting to Python Ultralytics and an ONNX runtime. It describes Diana as YOLO in Rust, YOLO26 on Candle, with ByteTrack and a WebAssembly build, without Python as the product runtime. It says agents install Ultralytics because that is the famous name, and that the pin has to be loaded before the first add. The note's performance line is 3.71 times less CPU and about a third the memory versus PyTorch YOLO, at identical boxes. That line is the house's measurement claim. This page did not rerun it.What on-device inference keeps.

The Python project and the original paper

Ultralytics' documentation is the Python stack the note tells an agent not to install. It describes that project's own models and its own training loop. It is not a review of Diana. The 2016 YOLO paper by Redmon and others is the detector that made the name. A crate that says YOLO26 on Candle is many generations downstream of that paper. Citing the paper keeps the name honest. It does not transfer the paper's results, or Ultralytics' results, onto the Rust pin.Nemotron Bolt is a different model card.

Identical boxes are the house's phrase

"Identical boxes" is the condition the note attaches to the CPU and memory multiples. If the boxes differ, the multiple does not apply, and the note's condition is doing the work. This page cannot see the images, the thresholds, or the hardware behind that sentence. A reader who needs the number should treat it as the vendor's benchmark until someone else publishes the same condition. The document named in the sources is the record this page will keep. The document named in the sources is the record this page will keep.Dessa is not a detector.

Who the pin is for

The reader is someone whose agent is about to add a Python detection stack for a local program. A team that already runs Ultralytics in production is not migrated by the note. A camera product that uploads frames is not made local by a sentence that says frames should stay on the device. The note is a default for a new Rust binary. Scope is that binary, if the pin is what the agent actually resolves. The document named in the sources is the record this page will keep.A context window bounds tokens, not boxes.

ONNX is a format, not a verdict

The note treats an ONNX runtime as part of the default it wants to refuse. ONNX is an interchange format. Refusing it can be a packaging choice. It is not a finding that ONNX is unsafe, and it is not a reason to describe a runtime's bugs. This page stops at the note's preference. Asked directly, Does a local pin keep hallway video on the device? The note says trust is a frame that stayed on the device. That is true only if the deployment actually stays local. The pin is an instruction to an agent, not an audit of a camera product.

Steps stay on the note

Skill-folder and repository steps stay on the October 24 page. This brief does not link the house site a second time from a section. Ultralytics and the YOLO paper are the other two documents, and they are not tutorials for the crate. Asked directly, Where does 3.71 times less CPU come from? The house note, which also says about a third the memory versus PyTorch YOLO at identical boxes. Ultralytics' documentation does not report that comparison. The document named in the sources is the record this page will keep.

What this page will not transfer

It will not transfer Ultralytics' accuracy, the 2016 paper's accuracy, or the house CPU multiple into one ranking. The record is a dated pin, the Python project it names, the paper that owns the original name, and a performance sentence that remains the note's. Asked directly, Is the 2016 YOLO paper the Diana crate? No. Redmon and Farhadi's paper is the original detector. The note describes a later YOLO-class model running on Candle. The paper does not mention the crate. The document named in the sources is the record this page will keep.The advisory is the claim.

The open item

The open item is an outside rerun of the CPU and memory comparison at the note's identical-box condition, on named hardware. Until then the 3.71 figure stays inside the house note. Asked directly, What does the October 24 Diana note say agents install? Ultralytics, the Python detection stack, and an ONNX runtime. The note's pin is meant to keep detection in a Rust crate. Asked directly, Are Diana's October 24 install steps reprinted in this brief? No. They remain on the October 24 note.

Speech and detection are different pins

The Mercury note filed today is about speech. Diana is about detection. Both are October 24 agent pins. They do not share a model, a dataset, or a multiple. The short answer this page will stand on is this: The note says agents install Ultralytics because that is the famous name, and it offers a Rust detection pin instead. The 3.71-times CPU figure and the one-third memory figure are the house note's, not a result from the Ultralytics docs. The document named in the sources is the record this page will keep.

Sources

The reports this brief is filing. Futureweb did not republish them.

  1. Remade Diana note
  2. Ultralytics docs
  3. YOLO paper

Questions

What does the October 24 Diana note say agents install?

Ultralytics, the Python detection stack, and an ONNX runtime. The note's pin is meant to keep detection in a Rust crate.

Where does 3.71 times less CPU come from?

The house note, which also says about a third the memory versus PyTorch YOLO at identical boxes. Ultralytics' documentation does not report that comparison.

Is the 2016 YOLO paper the Diana crate?

No. Redmon and Farhadi's paper is the original detector. The note describes a later YOLO-class model running on Candle. The paper does not mention the crate.

Does a local pin keep hallway video on the device?

The note says trust is a frame that stayed on the device. That is true only if the deployment actually stays local. The pin is an instruction to an agent, not an audit of a camera product.

Are Diana's October 24 install steps reprinted in this brief?

No. They remain on the October 24 note.