README.md

---
license: other
license_name: s-lab-1.0
license_link: https://github.com/pq-yang/MatAnyone/blob/main/LICENSE
library_name: coreml
pipeline_tag: image-segmentation
tags:
  - coreml
  - core-ml
  - ios
  - macos
  - apple
  - on-device
  - video-matting
  - alpha-matting
  - memory-network
  - arxiv:2501.14677
---

# MatAnyone — Core ML

*Video Matting, 2025*

Temporally consistent video matting. 5-model pipeline with memory propagation.

<p><img src="https://huggingface.co/mlboydaisuke/MatAnyone-CoreML/resolve/main/media/1418f0f55f.gif" alt="MatAnyone demo"></p>

Core ML conversion of [pq-yang/MatAnyone](https://github.com/pq-yang/MatAnyone) 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 | image segmentation |
| Upstream | [pq-yang/MatAnyone](https://github.com/pq-yang/MatAnyone) |
| Packages | 5 |
| Download size | 83 MB |
| Minimum iOS | 17.0 |
| Peak RAM | ~800 MB |

## Files

| File | Size | Compute units | SHA-256 |
|---|---:|---|---|
| `MatAnyone_encoder.mlpackage.zip` | 17 MB | `cpuAndGPU` | `97ffd6bc4611f9a3…` |
| `MatAnyone_mask_encoder.mlpackage.zip` | 16 MB | `cpuAndGPU` | `ba67559188ffc64d…` |
| `MatAnyone_read_first.mlpackage.zip` | 21 MB | `cpuOnly` | `34daf7227dbcec73…` |
| `MatAnyone_read.mlpackage.zip` | 21 MB | `cpuOnly` | `052e52c0ffb7ff9e…` |
| `MatAnyone_decoder.mlpackage.zip` | 8 MB | `cpuAndGPU` | `67136aa67000e604…` |
| **Total** | **83 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 "matanyone/*" --local-dir ./matanyone
unzip './matanyone/matanyone/*.zip' -d ./matanyone
```

## Use in Swift

```swift
import CoreML

let config = MLModelConfiguration()
config.computeUnits = .cpuAndGPU   // 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 MatAnyone_encoder(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 5 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/MatAnyoneDemo`](https://github.com/john-rocky/CoreML-Models/tree/master/sample_apps/MatAnyoneDemo), 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

- Script: [`convert_matanyone.py`](https://github.com/john-rocky/CoreML-Models/blob/master/conversion_scripts/convert_matanyone.py)
- 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: **S-Lab License 1.0**.
See [https://github.com/pq-yang/MatAnyone/blob/main/LICENSE](https://github.com/pq-yang/MatAnyone/blob/main/LICENSE).

> Non-commercial use only.

## Credits

- Upstream authors: [pq-yang/MatAnyone](https://github.com/pq-yang/MatAnyone), 2025
- Core ML conversion: john-rocky (Daisuke Majima)