Hi Stereolabs Support Team,
We’re seeing inconsistent behavior with ZED self-calibration on a ZED 2i using ZED SDK 5.1. The self-calibration sometimes succeeds and sometimes fails under what appear to be the same conditions, even when we follow the recommended guidelines (good illumination, unobstructed view, and textured elements in the scene).
System / setup details
- Camera: ZED 2i
- SDK: ZED SDK 5.1
- Runtime: we often run the camera for 2–3 hours in Neural Plus mode
- Environment: good illumination, clear FOV, textured scene
- Additional context: we also tried placing a checkerboard in front of the camera during self-calibration attempts
- Physical setup: two cameras facing each other, ~3.5 meters apart (if this matters for self-calibration stability)
Observed issue
- Self-calibration is not reliable: it alternates between “success” and “skipped” across runs, even when the scene and lighting appear consistent.
- When self-calibration fails, we consistently observe degraded depth quality (depth estimates are noticeably worse than when calibration succeeds).
Because the self-calibration algorithm isn’t open source, we’re trying to understand what practical conditions and constraints most affect its success and depth accuracy.
Questions (requesting detailed guidance)
-
Ideal conditions and scene requirements
- Beyond “good illumination + textured elements,” what are the most important scene conditions for consistent self-calibration success?
- Does a checkerboard help, or can it sometimes hurt (e.g., if it dominates the view or lacks natural texture)?
-
How to improve consistency
-
What actionable steps can we take to make self-calibration more consistently succeed?
- Recommended distance range to textured surfaces?
- Recommended motion pattern (if motion is required), and how much?
- Any recommended SDK parameters (e.g., depth mode, stabilization, camera control / auto-exposure on/off) that increase self-calibration stability?
-