Depth Pro 2 S
Smallest checkpoint for lower-memory and lower-latency inference.
Specifications
- Released
- Nov 19, 2026
- Inputs
- Images, Video
- Weights
- Open weights · Apple Machine Learning Research Model License
- Task
- Metric monocular depth estimation
- Published capabilities
- Smallest checkpoint for lower-memory and lower-latency inference.
- Model size
- 24M parameters, 97 MB weights
- Access
- Downloadable research weights and Apple’s reference inference code
- Result
- Per-pixel depth in meters at native resolution
- Operating limits
- Shares a separate 23.5M-parameter focal length estimator. The video pipeline supports buffered clips and causal streaming.
- Sources
- Model weights · Additional documentation