datagen_train.py

import numpy as np
import keras
import pickle



class DataGenerator(keras.utils.Sequence):
	'Generates data for Keras'
	def __init__(self, list_frames, labels, batch_size=32,n_classes=4, shuffle=True):
		'Initialization'
		self.batch_size = batch_size
		self.labels = labels
		self.list_frames = list_frames
		self.n_classes = n_classes
		self.shuffle = shuffle
		self.on_epoch_end()

	def __len__(self):
		'Denotes the number of batches per epoch'
		return int(np.floor(len(self.list_frames) / self.batch_size))

	def __getitem__(self, index):
		'Generate one batch of data'
		# Generate indexes of the batch
		indexes = self.indexes[index*self.batch_size:(index+1)*self.batch_size]

		# Find list of IDs
		list_frames_temp = [self.list_frames[k] for k in indexes]

		# Generate data
		X, y = self.__data_generation(list_frames_temp)

		return X, y

	def on_epoch_end(self):
		'Updates indexes after each epoch'
		self.indexes = np.arange(len(self.list_frames))
		if self.shuffle == True:
			np.random.shuffle(self.indexes)

	def __data_generation(self, list_frames_temp):
		'Generates data containing batch_size samples' # X : (n_samples, *dim, n_channels)
		# Initialization
		X = np.empty((self.batch_size, 299,299,3))
		y = np.empty((self.batch_size,4))


		# Generate data
		for i, video in enumerate(list_frames_temp):
			# Store sample
			# print(video[0])
			with open(video[0],'rb') as f1:		#for old model
				temp = pickle.load(f1)
				X[i,] = temp[int(video[1])]
			# X[i,] = np.load('bottleneck_data_sorted/'+video[0][16:]+video[1]+'.npy' , allow_pickle=True)
			#print(X.shape)
			# Store class
			y[i] = self.labels[video[0]]
			# print(X[i,] , y[i])
			#y = np.array([0,0,1,0])
			#print(y.shape)

		return X,y