| --- |
| license: apache-2.0 |
| base_model: depth-anything/da3-small |
| pipeline_tag: depth-estimation |
| library_name: coreai |
| tags: |
| - core-ai |
| - ios |
| - depth-estimation |
| --- |
| |
| # Depth Anything 3 small for Core AI |
|
|
| [`depth-anything/da3-small`](https://huggingface.co/depth-anything/da3-small) (35M parameters) exported to Apple Core AI with the [`apple/coreai-models`](https://github.com/apple/coreai-models/tree/main/models/depth-anything) recipe: |
|
|
| ```sh |
| uv run models/depth-anything/export.py |
| ``` |
|
|
| | Folder | Contents | Size | Platforms | |
| | ------------------- | --------------------------------------------------------------------------------------------------------------------------------------------- | ------ | ---------- | |
| | `da3-small.aimodel` | `main(image)`: `image` is `[1, 2, 3, 224, 224]` float32, two views of RGB normalized with the ImageNet mean and std. Returns `depth` and `depth_conf` (`[1, 2, 224, 224]`), `extrinsics` and `intrinsics` per view | 101 MB | iOS, macOS | |
|
|
| Depth is relative: smaller values are nearer. For a single photo, pass it as both views. |
|
|
| ## Use |
|
|
| With [`@react-native-ai/core-ai`](https://github.com/callstackincubator/ai/tree/main/packages/core-ai), download `da3-small.aimodel` into app storage, then: |
|
|
| ```ts |
| const { output } = await CoreAI.runTask( |
| 'depth', |
| { id: 'da3-small', sourceType: 'file', sourceUri: pathToModel }, |
| { imageUri }, |
| {} |
| ) |
| // output.image: base64 PNG, nearer is brighter. output.depth: the float32 map |
| ``` |
|
|
| ## License |
|
|
| Depth Anything 3 small is Apache-2.0 ([repository](https://github.com/ByteDance-Seed/Depth-Anything-3)). This repository only converts it to Core AI format. |
|
|
