Struct SimpleImageEncoder 

pub struct SimpleImageEncoder { /* private fields */ }

A lightweight multi-scale CNN encoder that produces feature maps at three spatial scales (1/4, 1/8, 1/16 of the input).

This replaces SAM’s ViT-H/ViT-L backbone with a practical pure-Rust convolutional architecture suitable for CPU inference.

Implementations

impl SimpleImageEncoder

pub fn new(config: &SAMConfig) -> Self

Build a new encoder from the given configuration.

pub fn encode(&self, image: &Array2<f64>) -> Result<Vec<Array2<f64>>>

Encode a single-channel image into multi-scale feature maps.

Arguments

• image – Grayscale image [H, W] with values in [0, 1].

Returns

A Vec of feature maps, one per encoder stage. Each feature map is [h_i * w_i, channels_i] (flattened spatial dims). The first entry corresponds to the coarsest (highest-level) features.

pub fn stage_channels(&self) -> Vec<usize>

Return the expected number of output channels at each stage.

Trait Implementations

impl Clone for SimpleImageEncoder

fn clone(&self) -> SimpleImageEncoder

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for SimpleImageEncoder

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

Formats the value using the given formatter.