Struct QuantizedConv2d 

pub struct QuantizedConv2d { /* private fields */ }

Quantized 2D Convolution Layer

Stores weights in INT8 format with per-channel scales. Performs computation in INT32 accumulator, then dequantizes to FP32.

Implementations

impl QuantizedConv2d

pub fn from_fp32(conv: &Conv2d, input_scale: f32, input_zero_point: i32) -> Self

Create from FP32 Conv2d with per-channel weight quantization

Arguments

• conv - FP32 convolution layer to quantize
• input_scale - Expected input activation scale
• input_zero_point - Expected input zero point

pub fn forward_int8( &self, input: &[u8], input_shape: &[usize], input_scale: f32, input_zero_point: u8, ) -> CnnResult<Tensor>

Forward pass with INT8 computation

Arguments

• input - Quantized u8 input tensor (NHWC layout)
• input_scale - Input quantization scale
• input_zero_point - Input quantization zero point

Returns

Dequantized FP32 output tensor

pub fn out_channels(&self) -> usize

Get number of output channels

pub fn in_channels(&self) -> usize

Get number of input channels

pub fn kernel_size(&self) -> usize

Get kernel size

pub fn stride(&self) -> usize

Get stride

pub fn padding(&self) -> usize

Get padding

Trait Implementations

impl Clone for QuantizedConv2d

fn clone(&self) -> QuantizedConv2d

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for QuantizedConv2d

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

Formats the value using the given formatter.