English

hbDNNTensorProperties

typedef struct {
  hbDNNTensorShape validShape;
  hbDNNTensorShape alignedShape;
  int32_t tensorLayout;
  int32_t tensorType;
  hbDNNQuantiShift shift;
  hbDNNQuantiScale scale;
  hbDNNQuantiType quantiType;
  int32_t quantizeAxis;
  int32_t alignedByteSize;
  int32_t stride[HB_DNN_TENSOR_MAX_DIMENSIONS];
} hbDNNTensorProperties;

Tensor information.

Member

Member Name	Description
`validShape`	Shape of the valid content of the tensor.
`alignedShape`	Shape of the aligned content of the tensor.
`tensorLayout`	Tensor layout form.
`tensorType`	Tensor type.
`shift`	Shift of quantization.
`scale`	Scale of quantization.
`quantiType`	Quantization type.
`quantizeAxis`	Index of the quantization axis, effective only when quantified by per-axis.
`alignedByteSize`	Memory size of the aligned content of the tensor.
`stride`	Step size of each dimension of validShape, measured in Bytes.

Note

If some dimensions in the validShape field are -1, it means that the validShape of the model is a dynamic input in this dimension, and you need to fill it in according to the actual input before creating a task.

The tensorLayout is compatible with J5, which is deprecated in J6.

The alignedShape is compatible with J5, which is not recommended for J6, using stride is recommended for aligning sizes.

The quantizeAxis represents the dimension of the quantization axis index and is only effective during per-axis quantization. When the data is quantized with per-tensor, quantizeAxis is negative.

The tensor information obtained through the interface is required by the model, you can modify the corresponding tensor information according to the actual input, at present, you are only allowed to modify the information of stride and tensorType, and must meet the requirements.

stride:

If you prepare your input based on stride, you do not need to modify the stride.
If you prepare your input based on validShape, you need to modify the value of stride to be the step size of the jumps in each dimension of validShape, and the inference library will internally padding the data.
If some dimensions of the stride obtained from the model are -1, it means that the stride of the model is a dynamic input in this dimension, and you need to fill it in according to the actual input before creating a task. For the constraints of filling in, please refer to the Dynamic Input Introduction section.

tensorType:

When inferring the model with NV12 input, you can modify the tensorType of the tensor to HB_DNN_IMG_TYPE_NV12 or HB_DNN_IMG_TYPE_NV12_SEPARATE based on the actual situation.

The alignedByteSize represents the size of the memory space occupied in the tensor, which can generally be used to apply for input and output memory.

PTQ Conversion Tools

The hb_compile Tool

PTQ Conversion Steps

PTQ Conversion Samples

FAQ and Common Failure Resolutions

Appendix

Tutorial

Advanced Tutorial

API Reference

QAT

Model Export

Horizon Operator

Vision Process

Vision Algorithm

Sample

VP API Introduction

Data Structure

Function Interface

Model Inference Development

Model Inference API Introduction

Data Type and Data Structure

API Interface

Model Inference Tool Introduction

High Performance Library

High Performance Algorithm

HPL API Introduction

Data Structure

Functional Interface

UCP Common API Introduction

Data Structure

Functional Interface

UCP Performance Analysis Tools

Horizon Torch Samples

FRAMEWORK

TUTORIALS

EXAMPLES

API REFERENCE

Toolchain Operator Support Constraint List

hbDNNTensorProperties

The hb_compile Tool

QAT

Model Export

Horizon Operator

Vision Algorithm

Sample

VP API Introduction

Data Structure

Function Interface

Model Inference API Introduction

Data Type and Data Structure

API Interface

Model Inference Tool Introduction

High Performance Algorithm

HPL API Introduction

Data Structure

Functional Interface

Data Structure

Functional Interface

FRAMEWORK

TUTORIALS

EXAMPLES

API REFERENCE

#hbDNNTensorProperties

hbDNNTensorProperties