Struct PhysicalOutput 

pub struct PhysicalOutput {
    pub name: String,
    pub type_: Option<PhysicalType>,
    pub shape: Vec<usize>,
    pub dshape: Vec<(DimName, usize)>,
    pub dtype: DType,
    pub quantization: Option<Quantization>,
    pub stride: Option<Stride>,
    pub scale_index: Option<usize>,
    pub activation_applied: Option<Activation>,
    pub activation_required: Option<Activation>,
}

Physical output: a concrete tensor produced by the converter.

Physical outputs carry only tensor-level fields (dtype, quantization, stride, activation_applied/activation_required). Semantic fields live on the LogicalOutput parent.

Fields

name: String

Physical tensor name as produced by the converter. This name is used to bind the metadata to the tensor returned by the inference runtime.

type_: Option<PhysicalType>

Semantic type. Matches the parent’s type or declares a sub-split such as boxes_xy or boxes_wh. None marks the child as type-opaque: it still binds to a physical tensor by name/shape during merging, but is not used to disambiguate against typed siblings. Useful when a converter emits extra per-scale tensors the HAL has no semantic for but the user manages downstream.

shape: Vec<usize>

Physical tensor shape.

dshape: Vec<(DimName, usize)>

Named dimensions ordered to match shape. Disambiguates NHWC vs NCHW per-child rather than assuming a model-wide layout.

dtype: DType

Quantized data type.

quantization: Option<Quantization>

Quantization parameters. Always present in v2; null means float (no quantization).

stride: Option<Stride>

FPN stride. Present on per-scale splits; absent on channel sub-splits (e.g. boxes_xy/boxes_wh).

scale_index: Option<usize>

Zero-based index into the parent’s strides array. Used for parallel iteration with precomputed per-scale state.

activation_applied: Option<Activation>

Activation already applied by the NPU. The HAL must NOT re-apply an activation declared here (e.g. Hailo applies sigmoid to score tensors on-chip).

activation_required: Option<Activation>

Activation NOT yet applied. The HAL MUST apply the declared activation before consuming the tensor.

Trait Implementations

impl Clone for PhysicalOutput

fn clone(&self) -> PhysicalOutput

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for PhysicalOutput

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for PhysicalOutput

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl PartialEq for PhysicalOutput

fn eq(&self, other: &PhysicalOutput) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Serialize for PhysicalOutput

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl StructuralPartialEq for PhysicalOutput