--- title: DepthPro2 emoji: 📷 colorFrom: blue colorTo: pink sdk: gradio sdk_version: 6.32.0 python_version: "3.12" app_file: app.py pinned: false license: other license_name: apple-amlr short_description: Metric monocular depth from photos and videos startup_duration_timeout: 1h --- # Depth Pro 2 — metric monocular depth demo Gradio demo for [apple/DepthPro2](https://huggingface.co/apple/DepthPro2) ([code](https://github.com/apple-aiml-research/ml-depth-pro-2)): metric depth for every pixel from a single photo or a whole video, sharp at object edges and steady from frame to frame. Uses the **DepthPro2-H** weights (872M parameters) with the focal length taken from the photo's EXIF when present. The app follows the model card's quick start: weights created with `create_model_and_transforms(..., precision=torch.bfloat16)`, `model.infer(...)` for images and `model.infer_video(...)` over frames decoded with PyAV. The focal length comes from the photo's EXIF when present, otherwise the bundled focal estimator estimates it. The app is written for ZeroGPU (`@spaces.GPU`-decorated functions, weights moved to CUDA inside the call); on plain CPU hardware it still runs, just slowly.
Built with ML InternAdd a live-camera tab ›Swap in the L weights ›