mlboydaisuke/RF-DETR-Nano-CoreML

RF-DETR Nano — Core ML

Object Detection, 2025

End-to-end transformer detector. 384×384 input. 300 queries, 91 classes (COCO + background). No NMS needed.

Core ML conversion of roboflow/rf-detr for on-device inference on iPhone, iPad and Mac. Converted with coremltools; the packages are stateless, so all sequencing and buffering lives in your Swift code.

Task	object detection
Upstream	roboflow/rf-detr
Packages	1
Download size	95 MB
Minimum iOS	17.0
Peak RAM	~400 MB

Files

File	Size	Compute units	SHA-256
rfdetr_n_coco.mlpackage.zip	95 MB	all	3cac3793b97aa88d…
Total	95 MB

compute_units is not a suggestion -- it is the configuration the conversion was verified against. Moving a package to a different compute unit can silently change the numerics (FP16 attention overflow) or crash on the GPU.

Download

hf download mlboydaisuke/coreml-zoo --include "rfdetr/*" --local-dir ./rfdetr_n
unzip './rfdetr_n/rfdetr/*.zip' -d ./rfdetr_n

Use in Swift

import CoreML

let config = MLModelConfiguration()
config.computeUnits = .all   // as converted — see the table above

// Unzip the .mlpackage, drop it into your Xcode target and Xcode compiles it
// at build time:
let model = try rfdetr_n_coco(configuration: config)

// ...or compile a downloaded .mlpackage at runtime:
let compiled = try await MLModel.compileModel(at: mlpackageURL)
let model = try MLModel(contentsOf: compiled, configuration: config)

Demo

• Sample app — peaceofcake/DFINEDemo, a standalone iOS project.
• Models Zoo — this model is downloadable and runnable inside the Models Zoo app on the App Store, no build required.

Conversion

• Pitfalls hit during conversion (FP16 overflow, ANE buffer limits, stride handling): docs/coreml_conversion_notes.md
• Model index: CoreML-Models

License

The conversion inherits the upstream license: Apache-2.0.

Credits

• Upstream authors: roboflow/rf-detr, 2025
• Core ML conversion: john-rocky (Daisuke Majima)