Visiongraph

Always the latest (tracks the current branch):

https://github.com/going-haywire/haybale-visiongraph/blob/master/marketstall.toml

Frozen to this version:

https://github.com/going-haywire/haybale-visiongraph/blob/v0.0.43/marketstall.toml

Visiongraph is a computer-vision pipeline library designed to simplify the prototyping of image-based algorithms

Nodes

Vision

• Annotate Results — Draw estimator results (boxes / masks / poses) onto a frame
• Frame Event — Triggered when a camera frame is ready; exposes colour/depth/infrared streams
• Frame Info Display — Displays information about frames with live preview
• OAK-D Camera — Opens an OAK-D depth camera and emits colour/depth/infrared frame callbacks
• Object Detector — Detect objects in a frame (bounding box + class + score)
• Pose Estimator — Estimate human body pose (named joints + skeleton) per person
• Segmentation — Instance segmentation: detect objects and their pixel masks
• Tracker — Assign stable tracking ids to detections across frames
• Web Camera — Starts a webcam stream and emits frame callbacks

Types

• Callback Multiframe — Subscription for multi-frame streams
• Detection Result — Object detections: bounding box, class, score (and tracking id)
• Landmark Result — Landmark detections: a detection plus a set of landmark points
• Pose Result — Human pose: landmarks with named joints and skeleton connections
• Segmentation Result — Instance segmentation: detection plus a per-instance mask
• Vision Result — A list of estimator results (base type for all result kinds)

Widgets

• NumpyViewerWidget — Streaming video viewer for numpy arrays using custom StreamingViewer