用于图像训练的深度学习

Question

我尝试训练图像并预测其中的文本。但是当一起训练图像时，我遇到了数组错误。但是现在我正在训练一封信的每个图像，但我遇到了一些错误。图像生成器文件被添加到有助于创建图像并将其导入拟合生成器的位置。

错误：

Using TensorFlow backend.
WARNING: Logging before flag parsing goes to stderr.
W0826 09:18:45.040408  3428 deprecation_wrapper.py:119] From C:\Users\workspace\test\venv\lib\site-packages\keras\backend\tensorflow_backend.py:74: The name tf.get_default_graph is deprecated. Please use tf.compat.v1.get_default_graph instead.

W0826 09:18:45.056031  3428 deprecation_wrapper.py:119] From C:\Users\workspace\test\venv\lib\site-packages\keras\backend\tensorflow_backend.py:517: The name tf.placeholder is deprecated. Please use tf.compat.v1.placeholder instead.

W0826 09:18:45.071652  3428 deprecation_wrapper.py:119] From C:\Users\workspace\test\venv\lib\site-packages\keras\backend\tensorflow_backend.py:4138: The name tf.random_uniform is deprecated. Please use tf.random.uniform instead.

Traceback (most recent call last):
  File "C:/Users/workspace/test/killme.py", line 22, in <module>
    o2 = Reshape((len(string.ascii_uppercase), ), name="symbol_{}".format(i+1))(o)
  File "C:\Users\workspace\test\venv\lib\site-packages\keras\engine\base_layer.py", line 474, in __call__
    output_shape = self.compute_output_shape(input_shape)
  File "C:\Users\workspace\test\venv\lib\site-packages\keras\layers\core.py", line 398, in compute_output_shape
    input_shape[1:], self.target_shape)
  File "C:\Users\workspace\test\venv\lib\site-packages\keras\layers\core.py", line 386, in _fix_unknown_dimension
    raise ValueError(msg)
ValueError: total size of new array must be unchanged

代码：

from keras.models import Sequential, Input, Model
from keras.layers import Dense, Reshape
from keras.utils import to_categorical
from keras.layers.convolutional import Conv2D # to add convolutional layers
from keras.layers.convolutional import MaxPooling2D # to add pooling layers
from keras.layers import Flatten # to flatten data for fully connected layers
import string
from generators import ImageGenerator, BasicGenerator
from numpy import reshape
height=20
width=200
font_size=20

i1=Input(shape=(height, width, 1))
character_count=int(width / font_size)
outputs=[]
for i in range(character_count):
    o = Dense(len(string.ascii_uppercase), activation='relu')(i1)
    o2 = Reshape((len(string.ascii_uppercase), ), name="symbol_{}".format(i+1))(o)
    outputs.append(o2)

string_model = Model(inputs=i1, outputs=outputs)
string_model.layers[2].layer.trainable = False

generator = ImageGenerator(height, width, font_size, character_count)
string_model.compile(loss="categorical_crossentropy", optimizer="rmsprop", metrics=["categorical_accuracy"])
string_model.summary()
string_model.fit_generator(generator,epochs=10)

Answer 1

在将文本数据放入密集输出之前，您必须对其进行预处理。转换成vocab 会更好。创建一个CaptionGenerator 使其简单如下。

from vgg16 import VGG16
from keras.applications import inception_v3
import numpy as np
import pandas as pd
from keras.models import Sequential
from keras.layers import LSTM, Embedding, TimeDistributed, Dense, RepeatVector, Merge, Activation, Flatten
from keras.preprocessing import image, sequence
from keras.callbacks import ModelCheckpoint
import cPickle as pickle

EMBEDDING_DIM = 128


class CaptionGenerator():

    def __init__(self):
        self.max_cap_len = None
        self.vocab_size = None
        self.index_word = None
        self.word_index = None
        self.total_samples = None
        self.encoded_images = pickle.load( open( "encoded_images.p", "rb" ) )
        self.variable_initializer()

    def variable_initializer(self):
        df = pd.read_csv('Flickr8k_text/flickr_8k_train_dataset.txt', delimiter='\t')
        nb_samples = df.shape[0]
        iter = df.iterrows()
        caps = []
        for i in range(nb_samples):
            x = iter.next()
            caps.append(x[1][1])

        self.total_samples=0
        for text in caps:
            self.total_samples+=len(text.split())-1
        print "Total samples : "+str(self.total_samples)

        words = [txt.split() for txt in caps]
        unique = []
        for word in words:
            unique.extend(word)

        unique = list(set(unique))
        self.vocab_size = len(unique)
        self.word_index = {}
        self.index_word = {}
        for i, word in enumerate(unique):
            self.word_index[word]=i
            self.index_word[i]=word

        max_len = 0
        for caption in caps:
            if(len(caption.split()) > max_len):
                max_len = len(caption.split())
        self.max_cap_len = max_len
        print "Vocabulary size: "+str(self.vocab_size)
        print "Maximum caption length: "+str(self.max_cap_len)
        print "Variables initialization done!"


    def data_generator(self, batch_size = 32):
        partial_caps = []
        next_words = []
        images = []
        print "Generating data..."
        gen_count = 0
        df = pd.read_csv('Flickr8k_text/flickr_8k_train_dataset.txt', delimiter='\t')
        nb_samples = df.shape[0]
        iter = df.iterrows()
        caps = []
        imgs = []
        for i in range(nb_samples):
            x = iter.next()
            caps.append(x[1][1])
            imgs.append(x[1][0])


        total_count = 0
        while 1:
            image_counter = -1
            for text in caps:
                image_counter+=1
                current_image = self.encoded_images[imgs[image_counter]]
                for i in range(len(text.split())-1):
                    total_count+=1
                    partial = [self.word_index[txt] for txt in text.split()[:i+1]]
                    partial_caps.append(partial)
                    next = np.zeros(self.vocab_size)
                    next[self.word_index[text.split()[i+1]]] = 1
                    next_words.append(next)
                    images.append(current_image)

                    if total_count>=batch_size:
                        next_words = np.asarray(next_words)
                        images = np.asarray(images)
                        partial_caps = sequence.pad_sequences(partial_caps, maxlen=self.max_cap_len, padding='post')
                        total_count = 0
                        gen_count+=1
                        print "yielding count: "+str(gen_count)
                        yield [[images, partial_caps], next_words]
                        partial_caps = []
                        next_words = []
                        images = []

    def load_image(self, path):
        img = image.load_img(path, target_size=(224,224))
        x = image.img_to_array(img)
        return np.asarray(x)


    def create_model(self, ret_model = False):
        #base_model = VGG16(weights='imagenet', include_top=False, input_shape = (224, 224, 3))
        #base_model.trainable=False
        image_model = Sequential()
        #image_model.add(base_model)
        #image_model.add(Flatten())
        image_model.add(Dense(EMBEDDING_DIM, input_dim = 4096, activation='relu'))

        image_model.add(RepeatVector(self.max_cap_len))

        lang_model = Sequential()
        lang_model.add(Embedding(self.vocab_size, 256, input_length=self.max_cap_len))
        lang_model.add(LSTM(256,return_sequences=True))
        lang_model.add(TimeDistributed(Dense(EMBEDDING_DIM)))

        model = Sequential()
        model.add(Merge([image_model, lang_model], mode='concat'))
        model.add(LSTM(1000,return_sequences=False))
        model.add(Dense(self.vocab_size))
        model.add(Activation('softmax'))

        print "Model created!"

        if(ret_model==True):
            return model

        model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])
        return model

    def get_word(self,index):
        return self.index_word[index]

点击链接了解更多信息：：https://github.com/arjun-kava/caption_generator