Deep learning for image training - python-3.x

Question

I tried training images and predict the text in it. But when training images all together I was getting array error. But now I am training each image of a letter but I am getting some error. The image generator file is added where it helps to create the images and importing it to the fit generator.
Error:
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
Code:
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

You have to preprocess text data before just putting into the output dense. Converting into vocab would be a better idea. Create a CaptionGenerator to make it simple as follows.
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]
Follow Link For more info: : https://github.com/arjun-kava/caption_generator