Struct TensorAttr 

pub struct TensorAttr {
    pub index: u32,
    pub shape: Vec<u32>,
    pub n_elems: u32,
    pub size: u32,
    pub size_with_stride: u32,
    pub format: TensorFormat,
    pub data_type: TensorType,
    pub qnt_type: QuantType,
    pub zp: i32,
    pub scale: f32,
    pub name: String,
}

Metadata for a single tensor (input or output).

Contains everything you need to interpret the tensor data: shape, memory layout, data type, and quantization parameters.

Quantization fields

• zp (zero-point) and scale are used for INT8 affine dequantization: f32_value = (raw_i8 - zp) * scale
• These are set during model conversion and are different for each tensor.

Shape

For NC1HWC2 outputs (common on RV1106), the shape is [1, c1, H, W, c2]. For NHWC inputs, the shape is [1, H, W, C].

Fields

index: u32

Tensor index (0 for first input/output, 1 for second, etc.).

shape: Vec<u32>

Tensor dimensions. See TensorFormat for how to interpret them.

n_elems: u32

Total number of elements in the tensor.

size: u32

Size in bytes.

size_with_stride: u32

Size in bytes including stride padding (used for memory allocation).

format: TensorFormat

Memory layout of the tensor data.

data_type: TensorType

Element data type.

qnt_type: QuantType

Quantization method.

zp: i32

Quantization zero-point (for affine dequantization).

scale: f32

Quantization scale (for affine dequantization).

name: String

Human-readable tensor name from the model.

Trait Implementations

impl Clone for TensorAttr

fn clone(&self) -> TensorAttr

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for TensorAttr

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

Formats the value using the given formatter.

impl From<&RknnTensorAttr> for TensorAttr

Implements From for TensorAttr

fn from(raw: &RknnTensorAttr) -> Self

Converts to this type from the input type.