Status: Needs Review

This page has not been reviewed for accuracy and completeness. Content may be outdated or contain errors.

Plugin System¶

The cuvis-ai plugin system enables extending the framework with custom nodes and functionality without modifying the core codebase. Distribute your algorithms via Git, share with the community, and maintain independent versioning.

Quick Start¶

from cuvis_ai_core.utils.node_registry import NodeRegistry

registry = NodeRegistry()
registry.load_plugins("configs/plugins/trackeval.yaml")

HOTAMetricNode = registry.get("HOTAMetricNode", instance=registry)

Guides¶

Plugin System Overview

Architecture, manifest rules, path resolution, and loading behavior.
Plugin Development Guide

Create, test, package, and tag a cuvis-ai plugin for local or Git-based use.

Official Plugin Manifests¶

configs/plugins/adaclip.yaml: released AdaCLIP plugin manifest
configs/plugins/ultralytics.yaml: released Ultralytics YOLO26 plugin manifest pinned to v0.1.0
configs/plugins/deepeiou.yaml: released DeepEIoU plugin manifest pinned to v0.1.0
configs/plugins/trackeval.yaml: released TrackEval plugin manifest pinned to v0.1.0
configs/plugins/sam3.yaml: local SAM3 plugin manifest for a checkout at D:\code-repos\cuvis-ai-sam3\sam3-init

Official Plugins¶

cuvis-ai-adaclip - AdaCLIP vision-language anomaly detection
cuvis-ai-ultralytics - Ultralytics YOLO26 nodes for cuvis.ai detection and tracking pipelines
cuvis-ai-deepeiou - DeepEIoU tracking and optional ReID extractors for cuvis.ai tracking pipelines
cuvis-ai-trackeval - HOTA, CLEAR, and Identity tracking metrics
cuvis-ai-sam3 - local SAM3 tracking workflows and prompt propagation nodes

See Plugin Nodes for available plugins and loading examples.