File size: 5,848 Bytes
747451d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 | # /*---------------------------------------------------------------------------------------------
# * Copyright (c) 2022-2023 STMicroelectronics.
# * All rights reserved.
# *
# * This software is licensed under terms that can be found in the LICENSE file in
# * the root directory of this software component.
# * If no LICENSE file comes with this software, it is provided AS-IS.
# *--------------------------------------------------------------------------------------------*/
import numpy as np
import tensorflow as tf
def _apply_rescaling(dataset: tf.data.Dataset = None,
scale: float = None,
offset: float = None) -> tf.data.Dataset:
"""
Applies rescaling to a dataset using a tf.keras.Sequential model.
Args:
dataset (tf.data.Dataset): The dataset to be rescaled.
scale (float): The scaling factor.
offset (float): The offset factor.
Returns:
The rescaled dataset.
"""
# Define the rescaling model
rescaling = tf.keras.Sequential([
tf.keras.layers.Rescaling(scale, offset)
])
# Apply the rescaling to the dataset
rescaled_dataset = dataset.map(lambda x, y: (rescaling(x), y))
return rescaled_dataset
def _apply_normalization(dataset: tf.data.Dataset = None,
mean: tuple[float] = None,
std: tuple[float] = None) -> tf.data.Dataset:
"""
Applies normalization to a dataset using a tf.keras.Sequential model.
Args:
dataset (tf.data.Dataset): The dataset to be rescaled.
mean (float): The mean of the three channels.
std (float): The variance of the three channels.
Returns:
The rescaled dataset.
"""
# Define the rescaling model
if isinstance(std, float):
variance = pow(std, 2)
elif isinstance(std, list):
variance = [x ** 2 for x in std]
else:
variance = [x ** 2 for x in std]
normalization = tf.keras.Sequential([
tf.keras.layers.Normalization(mean=mean, variance=variance)
])
# Apply the rescaling to the dataset
normalized_dataset = dataset.map(lambda x, y: (normalization(x), y))
return normalized_dataset
def apply_rescaling_on_image(image: tf.Tensor = None,
scale: float = None,
offset: float = None) -> tf.Tensor:
"""
Applies rescaling to an image using a tf.keras.Sequential model.
Args:
image (tf.Tensor): The image to be rescaled.
scale (float): The scaling factor.
offset (float): The offset factor.
Returns:
The rescaled image.
"""
# Define the rescaling model
rescaling = tf.keras.Sequential([
tf.keras.layers.Rescaling(scale, offset)
])
# Apply the rescaling to the image
rescaled_image = rescaling(image)
return rescaled_image
def preprocessing(dataset: tf.data.Dataset = None,
scale: float = None,
offset: float = None,
mean: tuple[float] = None,
std: tuple[float] = None) -> tf.data.Dataset:
"""
Applies a full preprocessing to an image using a tf.keras.Sequential model.
Args:
image (tf.Tensor): The image to be rescaled.
scale (float): The scaling factor.
offset (float): The offset factor.
mean (float): The mean of the three channels.
variance (float): The variance of the three channels.
Returns:
The rescaled and normalized image.
"""
if dataset != None:
dataset = _apply_rescaling(dataset=dataset,
scale=scale,
offset=offset)
dataset = _apply_normalization(dataset=dataset,
mean=mean,
std=std)
return dataset
def preprocess_input(image: np.ndarray, input_details: dict) -> tf.Tensor:
"""
Preprocesses an input image according to input details.
Args:
image: Input image as a NumPy array.
input_details: Dictionary containing input details, including quantization and dtype.
Returns:
Preprocessed image as a TensorFlow tensor.
"""
image = tf.image.resize(image, (input_details['shape'][1], input_details['shape'][2]))
# Get the dimensions
if input_details['dtype'] in [np.uint8, np.int8]:
image_processed = (image / input_details['quantization'][0]) + input_details['quantization'][1]
image_processed = np.clip(np.round(image_processed), np.iinfo(input_details['dtype']).min,
np.iinfo(input_details['dtype']).max)
else:
image_processed = image
image_processed = tf.cast(image_processed, dtype=input_details['dtype'])
image_processed = tf.expand_dims(image_processed, 0)
return image_processed
def postprocess_output(output: np.ndarray, output_details: dict) -> np.ndarray:
"""
Postprocesses the model output to obtain the predicted label.
Args:
output (np.ndarray): The output tensor from the model.
output_details: Dictionary containing output details, including quantization and dtype.
Returns:
np.ndarray: The predicted label.
"""
if output_details['dtype'] in [np.uint8, np.int8]:
# Convert the output data to float32 data type and perform the inverse quantization operation
output_flp = (np.float32(output) - output_details['quantization'][1]) * output_details['quantization'][0]
else:
output_flp = output
if output.shape[1] > 1:
predicted_label = np.argmax(output_flp, axis=1)
else:
predicted_label = np.where(output_flp < 0.5, 0, 1)
return predicted_label
|