Alert Verification Performance

The following table presents average latency metrics alerts verified through VLMs.

Past Snippet: Refers to video clips that have already been processed by VST.

Real Time Alerts: Represents alerts processed immediately upon generation.

Concurrent Alerts: Captures performance when multiple alerts are processed simultaneously by the VLM.

Alert Verification Latency

	Past Snippet	Real Time Alerts	Concurrent Alerts
Kafka Queue	0.05s	0.06s	N/A
Worker Dispatch	0.20s	0.36s	N/A
VST Fetch	0.04s	3.72s	0.05s
VLM Inference	2.00s	2.14s	5.73s