问ResourceExhaustedError:当分配形状为[16，224，224，256]且类型为bool的张量时，OOM

EN

# Importing Libraries
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import tensorflow as tf
from tensorflow.keras.layers import Dense, Dropout, Input, InputLayer, Conv2D,UpSampling2D,DepthwiseConv2D
from tensorflow.keras.layers import Conv2D, Conv2DTranspose, Reshape, Flatten,LeakyReLU, Dropout,DepthwiseConv2D
from tensorflow.keras.layers import Flatten,MaxPooling2D,Conv2DTranspose, AveragePooling2D
from tensorflow.keras.applications.vgg16  import VGG16
from tensorflow.keras.models import Model,Sequential
from tensorflow.keras.optimizers import Adam
from tensorflow.keras import layers
from PIL import Image
from tensorflow.keras.utils import plot_model
import cv2

#load data
ab = np.load('ab1.npy')
gray = np.load('gray_scale.npy')
ab.shape #(10000, 224, 224, 2)

def get_rbg(gray_imgs, ab_imgs, n = 10):
    
    #create an empty array to store images
    img1 = np.zeros((n, 224, 224, 3))
    
    img1[:, :, :, 0] = gray_imgs[0:n:]
    img1[:, :, :, 1:] = ab_imgs[0:n:]
    
    #convert all the images to type unit8
    img1 = img1.astype("uint8")
    
    #create a new empty array
    imgs= []
    
    for i in range(0, n):
        imgs.append(cv2.cvtColor(img1[i], cv2.COLOR_LAB2RGB))
        
    #convert the image matrix into a numpy array
    imgs = np.array(imgs)
    
    return imgs
img_out = tf.keras.applications.vgg16.preprocess_input(get_rbg(gray_imgs = gray, ab_imgs = ab, n = 100))# 300

img_out.shape #(100, 224, 224, 3)

model6 = VGG16(weights='imagenet',include_top=False,input_shape=(224, 224, 3))
model = Sequential()
model.add(InputLayer(input_shape=(img_in.shape[1], img_in.shape[2], 3)))
model.add(Model(inputs=model6.inputs, outputs=model6.layers[-10].output))
model.add(UpSampling2D((2, 2)))
model.add(UpSampling2D((2, 2)))
model.add(DepthwiseConv2D(32, (2, 2), activation=tf.nn.relu, padding='same'))
model.add(UpSampling2D((2, 2)))
model.add(DepthwiseConv2D(32, (2, 2), activation=tf.nn.relu, padding='same'))
model.add(layers.ReLU(0.3))
model.add(layers.Dropout(0.4))
model.add(UpSampling2D((2, 2)))
model.add(UpSampling2D((2, 2)))
model.add(DepthwiseConv2D(2, (2, 2), activation=tf.nn.relu, padding='same'))
model.add(layers.ReLU(0.3))
model.add(layers.Dropout(0.2))
model.add(UpSampling2D((2, 2)))
model.add(layers.ReLU(0.3))
model.add(layers.Dropout(0.2))
model.add(AveragePooling2D(pool_size = (2, 2)))
model.add(layers.Dense(units=3))
print(model.summary())

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
functional_1 (Functional)    (None, 56, 56, 256)       1735488   
_________________________________________________________________
up_sampling2d (UpSampling2D) (None, 112, 112, 256)     0         
_________________________________________________________________
up_sampling2d_1 (UpSampling2 (None, 224, 224, 256)     0         
_________________________________________________________________
depthwise_conv2d (DepthwiseC (None, 112, 112, 256)     262400    
_________________________________________________________________
up_sampling2d_2 (UpSampling2 (None, 224, 224, 256)     0         
_________________________________________________________________
depthwise_conv2d_1 (Depthwis (None, 112, 112, 256)     262400    
_________________________________________________________________
re_lu (ReLU)                 (None, 112, 112, 256)     0         
_________________________________________________________________
dropout (Dropout)            (None, 112, 112, 256)     0         
_________________________________________________________________
up_sampling2d_3 (UpSampling2 (None, 224, 224, 256)     0         
_________________________________________________________________
up_sampling2d_4 (UpSampling2 (None, 448, 448, 256)     0         
_________________________________________________________________
depthwise_conv2d_2 (Depthwis (None, 224, 224, 256)     1280      
_________________________________________________________________
re_lu_1 (ReLU)               (None, 224, 224, 256)     0         
_________________________________________________________________
dropout_1 (Dropout)          (None, 224, 224, 256)     0         
_________________________________________________________________
up_sampling2d_5 (UpSampling2 (None, 448, 448, 256)     0         
_________________________________________________________________
re_lu_2 (ReLU)               (None, 448, 448, 256)     0         
_________________________________________________________________
dropout_2 (Dropout)          (None, 448, 448, 256)     0         
_________________________________________________________________
average_pooling2d (AveragePo (None, 224, 224, 256)     0         
_________________________________________________________________
dense (Dense)                (None, 224, 224, 3)       771       
=================================================================
Total params: 2,262,339
Trainable params: 2,262,339
Non-trainable params: 0
_________________________________________________________________
None
model.compile(optimizer = tf.keras.optimizers.Adam(),loss = 'mse',metrics=tf.keras.metrics.Accuracy())
#if you encounter an OOM error, reduce the batch_size to 8
model.fit(img_in, img_out, epochs =5, batch_size = 5)