Android NDK API Reference - NeuralNetworks - Enumerations - OperandCode - quantization (8/16 位量化)

旧城等待， 2022-09-16 06:13 129阅读 0赞

#  Android NDK API Reference - NeuralNetworks - Enumerations - OperandCode - quantization (8/16 位量化) #

**Android NDK API Reference**  
[https://developer.android.google.cn/ndk/reference][https_developer.android.google.cn_ndk_reference]

**NeuralNetworks**  
[https://developer.android.google.cn/ndk/reference/group/neural-networks][https_developer.android.google.cn_ndk_reference_group_neural-networks]

enumeration [ɪˌnjuːmə'reɪʃ(ə)n]：n. 列举，计算，详述，细目
    asymmetric [ˌeɪsɪˈmetrɪk]：adj. 不对称的，不对等的
    symmetric [sɪ'metrɪk]：adj. 对称的

## 1. OperandCode ##

OperandCode

Operand types.  
操作数类型。

The type of an operand in a model.  
模型中操作数的类型。

Types prefaced with `ANEURALNETWORKS_TENSOR_*` must be used for tensor data (i.e., tensors with at least one dimension). Types not prefaced by `ANEURALNETWORKS_TENSOR_*` represent scalar values and must have no dimensions.  
以 `ANEURALNETWORKS_TENSOR_*` 开头的类型必须用于张量数据 (即至少具有一维的张量)。没有以 `ANEURALNETWORKS_TENSOR_*` 开头的类型表示标量值并且必须没有维度。

Although we define many types, most operators accept just a few types. Most used are `ANEURALNETWORKS_TENSOR_FLOAT32`, `ANEURALNETWORKS_TENSOR_QUANT8_ASYMM`, and `ANEURALNETWORKS_INT32`.  
尽管我们定义了许多类型，但大多数运算符只接受少数类型。

Available since NNAPI feature level 1.  
自 NNAPI feature level 1 起可用。

### 1.1 `ANEURALNETWORKS_BOOL` ###

An 8 bit boolean scalar value.  
一个 8 位布尔标量值。

Values of this operand type are either true or false. A zero value represents false; any other value represents true.

Available since NNAPI feature level 3.

### 1.2 `ANEURALNETWORKS_FLOAT16` ###

An IEEE 754 16 bit floating point scalar value.  
一个 IEEE 754 16 位浮点标量值。

Available since NNAPI feature level 3.

### 1.3 `ANEURALNETWORKS_FLOAT32` ###

A 32 bit floating point scalar value.

### 1.4 `ANEURALNETWORKS_INT32` ###

A signed 32 bit integer scalar value.

### 1.5 `ANEURALNETWORKS_MODEL` ###

A reference to a model.  
对模型的引用。

`ANeuralNetworksModel_setOperandValueFromModel` must be used to set the value for an Operand of this type.  
必须使用 `ANeuralNetworksModel_setOperandValueFromModel` 来设置这种类型的操作数的值。

Available since NNAPI feature level 4.

### 1.6 `ANEURALNETWORKS_TENSOR_BOOL8` ###

A tensor of 8 bit boolean values.  
8 位布尔值的张量。

Values of this operand type are either true or false. A zero value represents false; any other value represents true.

Available since NNAPI feature level 3.

### 1.7 `ANEURALNETWORKS_TENSOR_FLOAT16` ###

A tensor of IEEE 754 16 bit floating point values.

Available since NNAPI feature level 3.

### 1.8 `ANEURALNETWORKS_TENSOR_FLOAT32` ###

A tensor of 32 bit floating point values.

### 1.9 `ANEURALNETWORKS_TENSOR_INT32` ###

A tensor of 32 bit integer values.

### 1.10 `ANEURALNETWORKS_TENSOR_QUANT16_ASYMM` ###

A tensor of 16 bit unsigned integers that represent real numbers.  
16 位无符号整数的张量表示实数。

Attached to this tensor are two numbers that can be used to convert the 16 bit integer to the real value and vice versa. These two numbers are:  
附加到此张量的是两个数字，可用于将 16 位整数转换为实数值，反之亦然。这两个数字是：

*  `scale`: a 32 bit floating point value greater than zero.
 *  `zeroPoint`: a 32 bit integer, in range `[0, 65535]`.

The formula is: `real_value = (integer_value - zeroPoint) * scale`.  
反量化 (dequantization)：`(integer_value - zeroPoint) * scale => real_value`

Available since NNAPI feature level 3.

### 1.11 `ANEURALNETWORKS_TENSOR_QUANT16_SYMM` ###

A tensor of 16 bit signed integers that represent real numbers.  
16 位有符号整数的张量表示实数。

Attached to this tensor is a number representing real value scale that is used to convert the 16 bit number to a real value in the following way: `realValue = integerValue * scale`.  
附加到这个张量的是一个代表 real value scale 的数字，用于通过以下方式将 16 位数字转换为实数值：`realValue = integerValue * scale`。

反量化 (dequantization)：`integerValue * scale => realValue`

`scale` is a 32 bit floating point with value greater than zero.

Available since NNAPI feature level 3.

### 1.12 `ANEURALNETWORKS_TENSOR_QUANT8_ASYMM` ###

A tensor of 8 bit unsigned integers that represent real numbers.  
8 位无符号整数的张量表示实数。

Attached to this tensor are two numbers that can be used to convert the 8 bit integer to the real value and vice versa. These two numbers are:  
附加到此张量的是两个数字，可用于将 8 位整数转换为实数值，反之亦然。这两个数字是：

*  `scale`: a 32 bit floating point value greater than zero.
 *  `zeroPoint`: a 32 bit integer, in range `[0, 255]`.

The formula is: `real_value = (integer_value - zeroPoint) * scale`.  
反量化 (dequantization)：`(integer_value - zeroPoint) * scale => real_value`

### 1.13 `ANEURALNETWORKS_TENSOR_QUANT8_ASYMM_SIGNED` ###

A tensor of 8 bit signed integers that represent real numbers.  
8 位有符号整数的张量表示实数。

*  `scale`: a 32 bit floating point value greater than zero.
 *  `zeroPoint`: a 32 bit integer, in range `[-128, 127]`.

The formula is: `real_value = (integer_value - zeroPoint) * scale`.  
反量化 (dequantization)：`(integer_value - zeroPoint) * scale => real_value`

Available since NNAPI feature level 4.

### 1.14 `ANEURALNETWORKS_TENSOR_QUANT8_SYMM` ###

A tensor of 8 bit signed integers that represent real numbers.  
8 位有符号整数的张量表示实数。

Attached to this tensor is a number representing real value scale that is used to convert the 8 bit number to a real value in the following way: `realValue = integerValue * scale`.  
附加到这个张量的是一个代表 real value scale 的数字，用于通过以下方式将 8 位数字转换为实数值：`realValue = integerValue * scale`。

反量化 (dequantization)：`integerValue * scale => realValue`

`scale` is a 32 bit floating point with value greater than zero.

Available since NNAPI feature level 3.

### 1.15 `ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL` ###

A tensor of 8 bit signed integers that represent real numbers.  
8 位有符号整数的张量表示实数。

This tensor is associated with additional fields that can be used to convert the 8 bit signed integer to the real value and vice versa. These fields are:  
此张量相关联的附加字段可用于将 8 位有符号整数转换为实数值，反之亦然。这些字段是：

*  `channelDim`: a 32 bit unsigned integer indicating channel dimension.
 *  `scales`: an array of positive 32 bit floating point values. The size of the scales array must be equal to `dimensions[channelDim]`.  
    一个由 32 位正浮点值组成的数组。

`ANeuralNetworksModel_setOperandSymmPerChannelQuantParams` must be used to set the parameters for an Operand of this type.  
必须使用 `ANeuralNetworksModel_setOperandSymmPerChannelQuantParams` 来设置此类操作数的参数。

The channel dimension of this tensor must not be unknown (`dimensions[channelDim] != 0`).  
这个张量的通道维度不能是未知的。

The formula is: `realValue[..., C, ...] = integerValue[..., C, ...] * scales[C]` where `C` is an index in the Channel dimension.

Available since NNAPI feature level 3.

### 1.16 `ANEURALNETWORKS_UINT32` ###

An unsigned 32 bit integer scalar value.

## References ##

[https://developer.android.google.cn/ndk/reference/group/neural-networks][https_developer.android.google.cn_ndk_reference_group_neural-networks]  
[https://tensorflow.google.cn/lite/performance/quantization\_spec][https_tensorflow.google.cn_lite_performance_quantization_spec]

[https_developer.android.google.cn_ndk_reference]: https://developer.android.google.cn/ndk/reference
[https_developer.android.google.cn_ndk_reference_group_neural-networks]: https://developer.android.google.cn/ndk/reference/group/neural-networks
[https_tensorflow.google.cn_lite_performance_quantization_spec]: https://tensorflow.google.cn/lite/performance/quantization_spec