Data Generation

Object Segmentation Data Generation Code

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42

class CustomDataGen(tf.keras.utils.Sequence): def __init__(self, filenames, batch_size, shuffle=True): self.filenames = filenames.copy() # name of files self.batch_size = batch_size # batch size self.shuffle = shuffle # shuffle data if true self.n = len(self.filenames) # number of files def on_epoch_end(self): # shuffle data after each epoch if self.shuffle: np.random.shuffle(self.filenames) # def __get_data(self, batches): # Generates data containing batch_size samples X_batch = [] # input list y_batch = [] # output list for filename in batches: # loop to create new batch image = cv2.imread("image/"+filename) # load image mask = cv2.imread("mask/"+filename) # load mask of image, for segmentation X_batch.append(image) y_batch.append(mask) return np.array(X_batch, dtype="float32"), np.array(y_batch, dtype="float32") def __getitem__(self, index): # get name of the files for current batch batches = self.filenames[index * self.batch_size:(index + 1) * self.batch_size] # load images from file name X, y = self.__get_data(batches) return X, y def __len__(self): # return number of batches return self.n // self.batch_size

1 2 3 4 5 6 7

test_split = 0.3 training_data = CustomDataGen(filenames[:int(samples*(1-test_split))], batch_size) test_data = CustomDataGen(filenames[int(samples*(1-test_split)):samples], batch_size)

Convolution Layer

Convolutional Layers are deep learning layers that are commonly used on images. But their purpose is to extract features from spatial data.

The convolution layer uses filters of certain width and height, these filters convolve on the whole image. In Convolution Layers, weights are in the form of 1D, 2D, or 3D matrix. This weight matrix is multiple with the images.

Convolution Layer Parameters

Filters: Convolution layers can have many filters which are specified by filters parameter in Keras. Filters have sizes as width and height. filter size is called the kernel.

The number of pixels that filter moves is called Stride, stride will have different values of each dimension, such as in 2D there will be different values for row and column. in 3D there will be 3rd value for depth.