| --- |
| tags: |
| - coreml |
| - image-to-image |
| - deepmosaics |
| - computer-vision |
| - vision |
| license: mit |
| --- |
| |
| # DeepMosaics - CoreML Version |
|
|
| This repository contains the Apple CoreML conversions of the pre-trained models from the original [DeepMosaics](https://github.com/HypoX64/DeepMosaics) project by HypoX64. |
|
|
| DeepMosaics is a powerful tool for automatic video and image processing. By converting these models to **CoreML**, iOS and macOS developers can easily integrate these capabilities directly into their Apple ecosystem applications, taking full advantage of on-device hardware acceleration (Apple Neural Engine and GPU). |
|
|
| ## 📦 Available Models |
|
|
|
|
| * `add_mosaic.mlmodel` - 用於為圖片/影片添加馬賽克 |
| * `clean_mosaic_HD.mlmodel` - 用於去除馬賽克 (HD 高清版) |
| * `clean_mosaic_video.mlmodel` - 用於去除影片中的馬賽克 |
|
|
| ## 🚀 How to Use (Swift) |
|
|
| 1. Download the desired `.mlmodel` or `.mlpackage` file from the **[Files and versions](https://huggingface.co/)** tab. |
| 2. Drag and drop the file into your **Xcode** project. |
| 3. Xcode will automatically generate the Swift class for the model. |
| 4. Use it in your code: |
|
|
| ```swift |
| import CoreML |
| import Vision |
| |
| // 1. 初始化模型 (請替換成你的模型名稱) |
| let model = try! clean_mosaic_HD(configuration: MLModelConfiguration()) |
| |
| // 2. 準備輸入 (請根據你的模型實際的輸入規格修改) |
| // 例如，如果模型吃的是 CVPixelBuffer： |
| // let input = clean_mosaic_HDInput(inputImage: pixelBuffer) |
| |
| // 3. 執行推論 |
| // let output = try! model.prediction(input: input) |
| |
| // 4. 取得結果 |
| // let resultImage = output.outputImage |
| ``` |
|
|
| > **Note:** Please refer to Xcode's CoreML model inspector to see the exact expected input and output shapes (e.g., Image size or MultiArray dimensions). |
|
|
| ## ⚖️ Acknowledgements & License |
|
|
| - Original Project: [HypoX64/DeepMosaics](https://github.com/HypoX64/DeepMosaics) |
| - All credit for the original architecture and pre-trained weights goes to the original authors. |
| - The models are provided under the same license as the original project. |
|
|
