Skip to content

ai-segment.md

Latest commit

 

History

History
68 lines (42 loc) · 2.27 KB

ai-segment.md

File metadata and controls

68 lines (42 loc) · 2.27 KB

API: AI background segmentation

Where you are: docs → reference → api → ai-segment Read this first: api.md See also: gpu.md · api/keying.md

TL;DR Four endpoints configure per-source AI-driven background segmentation (people/foreground detection) with transparent, blurred, or solid-color backgrounds. Requires a GPU build with TensorRT; these endpoints are only registered when the segmentation engine is available. In non-GPU builds the registration is a no-op.

Endpoints

PUT /api/sources/{source}/ai-segment

Purpose: enable / reconfigure AI segmentation for a source. Handler: control/api_ai_segment.go → (*API).handleEnableAISegment.

Request body (control.aiSegmentRequest):

{ "sensitivity": 0.7, "edgeSmooth": 0.5, "background": "blur:20" }
Field Type Range Meaning
sensitivity float 0.0-1.0 matte decision threshold
edgeSmooth float 0.0-1.0 feathering on matte edges
background string special "" or "transparent", "blur:N" (N=1-50), "color:RRGGBB"

Defaults: sensitivity 0.7, edgeSmooth 0.5, background "".

Response 200: full ControlRoomState.

Errors: 400 (out-of-range sensitivity / edgeSmooth, bad background syntax, invalid hex color).

GET /api/sources/{source}/ai-segment

Purpose: return the current config for a source, or null if no config is set. Handler: (*API).handleGetAISegment.

Response 200: internal.AISegmentConfig or null.

DELETE /api/sources/{source}/ai-segment

Purpose: disable AI segmentation for a source and free GPU resources. Handler: (*API).handleDisableAISegment.

Response 204.

GET /api/ai-segment/status

Purpose: return global availability + loaded model name. Handler: (*API).handleAISegmentStatus.

Response 200:

{ "available": true, "modelName": "rvm-mobilenetv3-1.0", "sources": { "cam1": { "enabled": true, ... } } }

Related docs