models 包

使用 Azure ML 硬件加速模型服务加速的模型的 Python 表示形式。

模块

accel_model

包含 HW 加速模型的抽象基类的模块。

doesnotexisterror

不存在错误。

utils

模型的实用程序 - 大多与预处理相关。

Densenet121

Densenet 的 Float-32 版本。

此模型采用 RGB 格式,且缩放系数为 0.017

QuantizedDensenet121

Densenet 的量化版本。

此模型采用 RGB 格式。

QuantizedResnet152

Renset-152 的量化版本。

QuantizedResnet50

Renset-50 的量化版本。

QuantizedSsdVgg

SSD-VGG 的量化版本。

此模型采用 RGB 格式。

QuantizedVgg16

VGG-16 的量化版本。

此模型采用 RGB 格式。

Resnet152

Resnet-152 的 Float-32 版本。

Resnet50

Resnet-50 的 Float-32 版本。

SsdVgg

SSD-VGG 的 Float-32 版本。

此模型采用 RGB 格式。

Vgg16

VGG-16 的 Float-32 版本。

此模型采用 RGB 格式。

函数

preprocess_array

创建一个采用图像字节数组并返回正则化图像的 tensorflow 操作。

preprocess_array(in_images, order='RGB', scaling_factor=1.0, output_height=224, output_width=224, preserve_aspect_ratio=True)

参数

in_images
必需

图像字节的 [?] 维张量。 (通常为占位符)

order
默认值: RGB

通道顺序 -“BGR”或“RGB”

scaling_factor
默认值: 1.0

通道值的乘数

output_height
默认值: 224

输出图像高度

output_width
默认值: 224

输出图像宽度

preserve_aspect_ratio
默认值: True

如果为 True,则在缩放时保留图像纵横比

返回

图像的 float32 像素值的 [?, output_height, output_width, 3] 维张量。