Module infer 

Ultralytics YOLO schema inference from raw model I/O signals.

Model export pipelines (ONNX, TFLite, CoreML) carry Ultralytics-authored metadata (class names, task, input size) alongside the tensor shapes and dtypes the runtime reports. This module turns that raw signal into a crate::schema::SchemaV2 the decoder can act on, without requiring a hand-written edgefirst.json for every Ultralytics export.

Structs

InferredSchema

Result of inferring a schema from ModelSignals.

ModelSignals

Raw model I/O signals: tensor shapes/dtypes plus unparsed metadata, as reported by the inference runtime that loaded the model.

TensorInfo

A single input or output tensor as reported by the inference runtime.

Enums

InferError

Errors that can occur while inferring an Ultralytics schema.

ModelSource

Container format of the model the signals were read from. Drives Ultralytics per-format conventions (box normalization, tensor order).

Functions

infer_ultralytics_schema

Infers an Ultralytics YOLO SchemaV2 from raw model I/O signals and metadata.