Skip to main content

Detection Blocks

Source: lib/workflow_blocks/detection.py

These blocks run object detection or segmentation models against input images.

YOLO Detection​

Runs a YOLO26 model (the active one from the registry, or a specific version) against an input image and emits bounding boxes.

Inputs: image: ndarray (H, W, 3) Outputs: detections: list[Detection] Params:

  • model_id — model registry UUID, or "active" for the default
  • confidence — minimum confidence (default 0.25)
  • iou — NMS IoU threshold (default 0.45)
  • classes — optional class allowlist

SAM Segmentation​

Runs SAM 3 against an image with a text or visual prompt, returns masks.

Inputs: image, optional prompt_text or prompt_box Outputs: masks: list[Mask] Params:

  • prompts — list of text prompts
  • threshold — mask confidence cutoff
  • resolution — input resize before inference

SAM Video Track​

Runs SAM 3 against a stream of frames and tracks instances across them.

Inputs: frames: iterator[ndarray] Outputs: tracks: list[Track] Params: prompts, tracking confidence, max gap frames