Frame rate and result age
FPS describes how many frames are processed over time; latency is the time from acquisition until a result is usable. A long queue can deliver old results despite high throughput. State measurement start and end points explicitly. Observe the latency distribution and unusual cases alongside the average.
Queue policy depends on the task
A live display may prefer the newest frame; product inspection may not tolerate skipping an image. Define required frames through product triggers and process needs. If processing falls behind, expose it as an explicit condition. Silent frame loss must not turn into an incorrect acceptance decision.
An example timing budget
In a hypothetical line, an item travels 0.5 metres from camera to reject point at 1 metre per second, leaving roughly 0.5 seconds. The model cannot use all that time: communication, actuator response and margin also consume it. This is not SOPON performance data; establish a real budget from site distance and timing measurements.