Depth Pro 2: Sharp and Temporally Consistent Metric Monocular Depth

What happened
A team of researchers introduced Depth Pro 2, an AI model designed to estimate dense, metric depth from both images and videos at their original resolution. The model processes input from a single photo, a video, or a live camera feed, producing depth maps that are both sharp at object boundaries and consistent across frames. Depth Pro 2 was released along with its code and model weights, making it available for public use. The model is positioned as an improvement over previous monocular depth estimation methods in terms of accuracy, sharpness, and temporal consistency.
Key details
- The model can handle single images, entire videos, or live feeds at full input resolution.
- It produces depth maps that are metrically accurate, with sharp object edges and stable results across frames.
- A family of models and their code and weights have been released for public use.
Why it matters
Developers and researchers working with computer vision can now generate high-quality, metrically accurate depth maps from ordinary cameras without sacrificing sharpness or temporal consistency. This capability reduces the need for specialized hardware or complex multi-frame processing, enabling more accessible and scalable depth estimation for applications such as robotics, augmented reality, and 3D reconstruction.
