| --- |
| license: mit |
| pipeline_tag: image-to-image |
| library_name: coreai |
| tags: |
| - core-ai |
| - ios |
| - super-resolution |
| --- |
| |
| # EDSR x2 for Core AI |
|
|
| EDSR (`edsr_r16f64_x2`, 1.5M parameters) from [torchSR](https://github.com/Coloquinte/torchSR), exported to Apple Core AI with the [`apple/coreai-models`](https://github.com/apple/coreai-models/tree/main/models/edsr) recipe: |
|
|
| ```sh |
| uv run models/edsr/export.py --dtype float16 --dynamic |
| ``` |
|
|
| | Folder | Contents | Size | Platforms | |
| | ----------------- | -------------------------------------------------------------- | ------ | ---------- | |
| | `edsr-x2.aimodel` | `main(x)`: `x` is `[batch, 3, height, width]` RGB in 0...1, float16, batch 1–64, height and width 8–256. Returns the image at twice the size | 2.6 MB | iOS, macOS | |
|
|
| Images larger than 256 pixels are upscaled in tiles. |
|
|
| ## Use |
|
|
| With [`@react-native-ai/core-ai`](https://github.com/callstackincubator/ai/tree/main/packages/core-ai), download `edsr-x2.aimodel` into app storage, then call it with `CoreAI.runRawFunction`, or through `CoreAI.runTask('super-resolution', ...)`, which tiles and converts images for you. |
|
|
| ## License |
|
|
| torchSR and its EDSR weights are MIT licensed ([EDSR paper](https://arxiv.org/abs/1707.02921)). This repository only converts them to Core AI format. |
|
|
