Struct SemanticConfig 

pub struct SemanticConfig { /* private fields */ }

Static description of a semantic output, usually built from the model’s edgefirst.json by SemanticConfig::from_logical.

Implementations

impl SemanticConfig

pub fn scores( shape: &[usize], class_axis: usize, height_axis: usize, width_axis: usize, ) -> DecoderResult<Self>

Score map of shape, with the class, height and width axes at the given positions. Every other axis must have size 1.

Defaults: bilinear upsampling with align_corners = false, DecodeMode::Accurate, no activation applied, not quantized.

pub fn class_map( shape: &[usize], height_axis: usize, width_axis: usize, ) -> DecoderResult<Self>

Class map of shape with the height and width axes at the given positions. Every other axis must have size 1.

pub fn from_logical(logical: &LogicalOutput) -> DecoderResult<Self>

Builds the configuration of a segmentation or masks logical output with decoder: semantic, or of a ModelPack segmentation output, which is NHWC and decoded in DecodeMode::Fast unless it says otherwise.

pub fn with_upsample(self, upsample: Upsample) -> DecoderResult<Self>

Sets the score-map upsampling. Bicubic is not supported.

pub fn with_mode(self, mode: DecodeMode) -> Self

Sets the default decode mode.

pub fn with_threshold(self, threshold: f32) -> Self

Sets the single-channel foreground threshold, in the space of the scores after activation.

pub fn with_activation(self, activation: Activation) -> DecoderResult<Self>

Declares an activation the model already applied to the scores.

Errors

See Self::with_activation_required.

pub fn with_activation_required( self, activation: Activation, ) -> DecoderResult<Self>

Declares the activation the scores still need. It is applied after resampling, as reference pipelines do, so it leaves the class map unchanged and sets how decode_semantic computes confidence.

Errors

DecoderError::InvalidConfig when an activation is both applied and required, or a single channel declares softmax; DecoderError::NotSupported for tanh.

pub fn with_quantization(self, quantization: Quantization) -> Self

Quantization of integer scores, used when the tensor carries none.

pub fn with_name(self, name: impl Into<String>) -> Self

Names the output so select_output prefers the tensor of that name.

pub fn with_position(self, index: usize, count: usize) -> Self

Records the output’s index among the model’s count outputs, which select_output uses when several outputs share its shape.

pub fn input(&self) -> SemanticInput

What the output holds.

pub fn name(&self) -> Option<&str>

Output name, when known.

pub fn shape(&self) -> &[usize]

Expected tensor shape.

pub fn layout(&self) -> SemanticLayout

Axis positions in Self::shape.

pub fn classes(&self) -> usize

Number of classes the decoded class map can contain: the channel count, 2 for a single thresholded channel, 0 when unknown (a class map).

pub fn height(&self) -> usize

Score-map height.

pub fn width(&self) -> usize

Score-map width.

pub fn upsample(&self) -> Upsample

Score-map upsampling.

pub fn mode(&self) -> DecodeMode

Default decode mode.

pub fn threshold(&self) -> f32

Single-channel foreground threshold in the space of the scores: the configured value, else 0.5 for probabilities and 0.0 for logits.

pub fn matches(&self, tensor: &TensorDyn) -> bool

Whether tensor has this output’s shape, ignoring axes of size 1.

Trait Implementations

impl Clone for SemanticConfig

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for SemanticConfig

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

Formats the value using the given formatter.

impl PartialEq for SemanticConfig

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

Equality operator ==.

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

Inequality operator !=.

impl StructuralPartialEq for SemanticConfig