File size: 3,462 Bytes
7e63d1e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
---
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)