| --- |
| license: agpl-3.0 |
| library_name: coreml |
| pipeline_tag: zero-shot-object-detection |
| tags: |
| - coreml |
| - core-ml |
| - ios |
| - macos |
| - apple |
| - on-device |
| - yolo |
| - open-vocabulary |
| - instance-segmentation |
| - mobileclip |
| - zero-shot |
| - arxiv:2503.07465 |
| --- |
| |
| # YOLOE-S — Core ML |
|
|
| *Tsinghua, 2025* |
|
|
| Open-vocabulary detection **and** instance segmentation. Type any text — "person", "forklift", "coffee cup" — and get boxes plus masks, with no fixed class list. |
|
|
| Unlike a baked-in text head, the detector emits a per-anchor **region embedding** before the class logits and the region-text similarity is computed on the client. The image branch never sees the text, so changing the query does not re-run the detector — only a cheap matmul against cached text embeddings. |
|
|
| Core ML conversion of [THU-MIG/yoloe](https://github.com/THU-MIG/yoloe) 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 | zero shot object detection | |
| | Upstream | [THU-MIG/yoloe](https://github.com/THU-MIG/yoloe) | |
| | Packages | 3 | |
| | Download size | 133 MB | |
| | Minimum iOS | 17.0 | |
|
|
| ## Files |
|
|
| | File | Size | Compute units | SHA-256 | |
| |---|---:|---|---| |
| | `yoloe_detector_s.mlpackage.zip` | 18 MB | `all` | - | |
| | `mobileclip_blt_text.mlpackage.zip` | 112 MB | `all` | - | |
| | `reprta_s.mlpackage.zip` | 4 MB | `all` | - | |
| | **Total** | **133 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 |
|
|
| ```bash |
| hf download mlboydaisuke/coreml-zoo --include "yoloe/*" --local-dir ./yoloe |
| unzip './yoloe/yoloe/*.zip' -d ./yoloe |
| ``` |
|
|
| ## Use in Swift |
|
|
| ```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 yoloe_detector_s(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) |
| ``` |
|
|
| > This model is split into 3 Core ML packages that are driven in sequence from Swift. Load them one at a time, copy the outputs out of the `MLMultiArray` buffers and release each model before loading the next — two large Core ML models resident at once will OOM on an iPhone. |
|
|
| ## Demo |
|
|
| - **Sample app** — [`sample_apps/YOLOEDemo`](https://github.com/john-rocky/CoreML-Models/tree/master/sample_apps/YOLOEDemo), a standalone SwiftUI project. |
| - **Models Zoo** — this model is downloadable and runnable inside the [Models Zoo app](https://apps.apple.com/app/id6762083207) on the App Store, no build required. |
|
|
| ## Conversion |
|
|
| - Pitfalls hit during conversion (FP16 overflow, ANE buffer limits, stride handling): [`docs/coreml_conversion_notes.md`](https://github.com/john-rocky/CoreML-Models/blob/master/docs/coreml_conversion_notes.md) |
| - Model index: [CoreML-Models](https://github.com/john-rocky/CoreML-Models) |
|
|
| ## License |
|
|
| The conversion inherits the upstream license: **AGPL-3.0**. |
|
|
| ## Credits |
|
|
| - Upstream authors: [THU-MIG/yoloe](https://github.com/THU-MIG/yoloe), 2025 |
| - Core ML conversion: john-rocky (Daisuke Majima) |
|
|
