Segment Image (Triton)

How it fits

Typical backends

Caveats

• I/O contractThe output arm is a single downstream-resolved choice fixed at deploy time; every arm describes the same detected instances, so masks, boxes and classes stay aligned.
• Hard constraintThe model must already be served by the Triton inference service this component depends on; the component sends requests rather than loading weights itself, and the output tensor names configured here must match the served model's tensor names or the call fails.
• AccuracyEvery preprocessing setting must match the served model's expected input; mismatched mean, std, rescale_factor or color_model silently produce meaningless detections.
• Hard constraintA class id listed in classes that exceeds num_classes is rejected at startup, since the served network cannot emit that category.
• Parameter interactionresize_detections_height and resize_detections_width default to using the input size; non-zero values report boxes at that resolution instead.
• State lifetimeAll configuration is captured once at startup; changing any setting requires a redeploy.