pub struct ModelInfo {
pub id: String,
pub kind: ModelKind,
pub execution: ModelExecution,
pub backend_loaded: bool,
pub output_classes: Vec<String>,
pub fps: f32,
pub latency_ms: f32,
pub is_inference_capable: bool,
}Expand description
A registered model’s info, for the engine’s list-models read-back (the
methods::LIST_MODELS reply). Lets the GCS vision hub show every model
loaded on the drone — its task, how it runs, whether the engine loaded a
backend for it — not only the models currently publishing detections.
Fields§
§id: String§kind: ModelKind§execution: ModelExecution§backend_loaded: boolTrue when the engine holds a loaded backend model for it (an engine-run model whose file loaded); false for a plugin-side or failed-to-load model.
output_classes: Vec<String>§fps: f32Measured inference rate in Hz (an EWMA over inter-inference intervals).
0.0 when the model has not run yet — the honest “no data” reading, not a
claim of zero throughput. Additive: an older payload without it decodes to
0.0.
latency_ms: f32Measured per-inference latency in ms (an EWMA). 0.0 until the model runs.
is_inference_capable: boolWhether this model actually runs a real backend (an engine-run model whose
file loaded on an inference-capable backend). A mock-backend engine reports
false so a status surface never presents a placeholder as a working
detector. Additive; defaults false.