| |
| |
| |
| |
|
|
| from __future__ import annotations |
|
|
| import numpy as np |
|
|
| from qai_hub_models.utils.asset_loaders import load_image |
|
|
| from .app import DepthProApp |
| from .demo import INPUT_IMAGE_ADDRESS |
| from .demo import main as demo_main |
| from .model import DepthPro |
|
|
|
|
| def test_task() -> None: |
| """Run torch DepthPro end-to-end on the sample fixture. |
| |
| Sanity-checks that the pipeline resolves the HF weights, produces a |
| depth map at the network's native 1536x1536 unpadded to the original |
| input resolution, and yields a plausible field of view (roughly the |
| range Apple demos on natural imagery, 30-100 degrees). |
| """ |
| model = DepthPro.from_pretrained() |
| (_, _, height, width) = model.get_input_spec()["image"][0] |
| app = DepthProApp(model, height, width) |
| image = load_image(INPUT_IMAGE_ADDRESS) |
| prediction = app.estimate_depth(image) |
|
|
| assert prediction.depth.ndim == 2 |
| assert prediction.depth.shape == (image.size[1], image.size[0]) |
| assert np.all(np.isfinite(prediction.depth)) |
| assert prediction.depth.min() > 0 |
|
|
| assert 10.0 < prediction.field_of_view < 170.0 |
| assert prediction.focal_length_px > 0 |
|
|
| assert prediction.heatmap.size == image.size |
|
|
|
|
| def test_demo() -> None: |
| demo_main(is_test=True) |
|
|
