Monocular Stream Depth Estimation

Estimates a relative depth map from each frame of a live monocular camera stream and emits a synchronized depth output stream for live inspection.

Input

One image stream with a camera calibration data component.

Output

A child depth stream on the input camera stream. Each processed frame produces a depth map paired with the original color image so the display can show a live point-cloud preview.

Description

This algorithm applies the same monocular depth backend as Monocular Depth Estimation, frame-by-frame on a live feed. Use that algorithm for offline image sets; use this one when the camera is streaming and you need continuous depth aligned with the color frames.

Press Compute to attach the output stream. Stream playback and focus controls follow the usual stream workflow.

Depth backend

By default each frame is processed with Monocular Depth Estimation. Its properties (including Depth estimator name and Export point clouds) apply per frame. Depth remains relative (non-metric); calibration enables correct pinhole unprojection for preview only.

Processing

Reconstruction may lag behind the camera frame rate:

Queue size — number of frames buffered while waiting for depth estimation. When the queue is full, new frames are dropped.

Force blocking — when enabled, each frame is processed on the input thread instead of a background queue (higher latency, simpler timing).

Process on GPU — when enabled, incoming images are uploaded to the GPU before processing (required for some GPU-backed depth backends).