230523 CNN, Lightening

레이어 추가 및 이미지 차원 추가 여러가지 방법들

import tensorflow as tf

import numpy as np

from keras.api._v2.keras import activations

from keras.layers import Reshape

from tensorflow.keras.models import Sequential

from tensorflow.keras.layers import Dense, Flatten, Dropout, MaxPooling2D

from keras.engine.input_layer import InputLayer

tf.keras.datasets.mnist.load_data()

(x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()

x_train = x_train / 255.

x_test = x_test / 255.

# 레이어추가 요런 방법도 있다

model = Sequential()

model.add(InputLayer(input_shape=(28,28)))

model.add(Reshape(target_shape=(28,28,1)))

model.add(Conv2D(

    filters=12,

    kernel_size=(3,3),

    activation="relu"

))

model.add(MaxPooling2D(pool_size=(2,2)))

model.add(Flatten())

model.add(Dense(256, activation="relu"))

model.add(Dense(10, activation="softmax"))

# 이미지 차원 추가의 여러가지 방법

# 기본

x_train = x_train.reshape( (x_train.shape[0] , 28, 28, 1) )

x_test  = x_test.reshape( (x_test.shape[0] , 28, 28, 1) )

# 기타 다른 방법들. 출력결과 => (60000, 28, 28, 1)

[*x_train.shape, 1]

np.expand_dims(x_train, -1).shape

tf.expand_dims(x_train, -1).shape

x_train[..., tf.newaxis].shape

 

 

 

모델을 경량화하는 방법들

 

converter (데이터 타입 변경)

converter = tf.lite.TFLiteConverter.from_keras_model(model)

converter.optimizations = [tf.lite.Optimize.DEFAULT]

tflite_model = converter.convert()

# 모델의 input_type, output_type 출력해주는 함수

def check_in_out_type(model_content):

  interpreter = tf.lite.Interpreter(model_content=model_content)

  def get_details(func):

    return func()[0]["dtype"]

  input_type = interpreter.get_input_details()[0]["dtype"]

  output_type = interpreter.get_output_details()[0]["dtype"]

  print("input_type, output_type: ", input_type, output_type)

# 데이터타입 float32

def representative_data_gen():

  for elem in tf.data.Dataset.from_tensor_slices(

      x_train.astype(np.float32)

  ).batch(1).take(100):

      yield[elem]

# 데이터타입 uint8 변경용 컨버터

converter = tf.lite.TFLiteConverter.from_keras_model(model)

converter.optimizations = [tf.lite.Optimize.DEFAULT]

converter.representative_dataset = representative_data_gen

converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS_INT8]

converter.inference_input_type = tf.uint8

converter.inference_output_type = tf.uint8

tflite_model_quant = converter.convert()

# 데이터타입 float16 변경용 컨버터

converter = tf.lite.TFLiteConverter.from_keras_model(model)

converter.optimizations = [tf.lite.Optimize.DEFAULT]

converter.target_spec.supported_types = [tf.float16]

tflite_model_fp16 = converter.convert()

# 모델 저장하기

import pathlib

tflite_model_dir = pathlib.Path("/tmp/mnist_tflite_models/")

tflite_model_dir.mkdir(exist_ok=True, parents=True)

# Float32 Model

tflite_model_file = tflite_model_dir/"mnist_model.tflite"

tflite_model_file.write_bytes(tflite_model)

# Int8 Model

tflite_model_quant_file = tflite_model_dir/"tflite_model_quant"

tflite_model_quant_file.write_bytes(tflite_model_quant)

# Float16 Model

tflite_model_fp16_file = tflite_model_dir/"tflite_model_fp16"

tflite_model_fp16_file.write_bytes(tflite_model_fp16)