AI GitHub
Depth Anything V2 for monocular depth estimation
Estimate scene depth from a single image with published checkpoints.
Key Takeaways
- The model predicts depth cues without stereo input.
- Choose a checkpoint based on whether the app needs relative ordering or metric distances.
- Try a hallway and a reflective scene and inspect ordering around object boundaries.
- Relative depth values do not directly correspond to meters.
How it works
The model predicts depth cues without stereo input. The repository distinguishes relative depth models from documented metric-depth variants.
What you need
Choose a checkpoint based on whether the app needs relative ordering or metric distances. Keep camera and image preparation details with any reference measurements.
What to try first
Try a hallway and a reflective scene and inspect ordering around object boundaries. If you need metric distance, compare the appropriate model against a measured reference.
Use a reference where two objects have similar texture but different distances. This makes the distinction between approximate ordering and useful metric estimates visible.
Where to look for failures
Relative depth values do not directly correspond to meters. Lighting, reflections and domain shifts can affect either variant.
Related Franklin pages
Depth, object boxes and masks answer different questions about the same image. Use a few boundary cases to decide which output is needed and how errors will be caught.
- Generate object masks from visual prompts and propagate them through video. SAM 2 for prompted image and video segmentation
- Find image regions that match natural-language phrases. Grounding DINO for text-guided object detection
For an overview of the types of predictions vision systems produce, see Franklin’s glossary. Computer vision
Project documentation
Try it for your workflow
Explore the project’s documentation for setup and usage, or save this page to a collection for later.
Read the project documentation ↗