padding=padding_type) def Normalize_1D(images): images_flatten = Images.Flatten(images) return Image.Normalize(images_flatten) def Normalize_2D(images): return Image.Normalize(images) X_grey = Image.LoadTrainingGreyImage(dataset_size, './Training/X2_grey/') Y_grey = Image.LoadTrainingGreyImage(dataset_size, './Training/HR_grey/') X_norm = Image.ExpandDims(Normalize_2D(X_grey)) Y_norm = Image.ExpandDims(Normalize_2D(Y_grey)) X_3dims = Image.ExpandDims(X_grey) Y_3dims = Image.ExpandDims(Y_grey) imsave('tmp.png', X_grey[0].astype(int)) imsave('tmp1.png', Y_grey[0].astype(int)) X_train = tf.placeholder(tf.float32) Y_train = tf.placeholder(tf.float32) # 1st Layer - Features Extraction print('1st layer') weight_conv1 = weight_variable( [filter_size_conv1, filter_size_conv1, n_channels_conv1, n_filters_conv1])
path_local = './Training/' # Hyper params learning_rate = 0.0002 epochs = 500 batch_size = 40 dataset_size = 100 X_grey = Image.LoadTrainingGreyImage(dataset_size, path_gpu + 'X2_grey/') Y_grey = Image.LoadTrainingGreyImage(dataset_size, path_gpu + 'HR_grey/') print('finish reading') X_norm = Image.Normalize(X_grey) Y_norm = Image.Normalize(Y_grey) print('finish Normalize') X_cropped = Image.Segment(X_norm, 256) Y_cropped = Image.Segment(Y_norm, 512) print('finish cropping') X_final = Image.ExpandDims(X_cropped) Y_final = Image.ExpandDims(Y_cropped) print('finish ExpandDims') SuperResolution.Train(X_final, Y_final, Image.ExpandDims(X_norm), learning_rate, epochs, batch_size)
import tensorflow as tf import numpy as np from skimage.io import imread, imsave from Image import Image import SuperResolution # np.set_printoptions(threshold=np.nan) batch_size = 30 dataset_size = 1 model_path = './Models/128_32_300_30_100_ssim/' X_grey = Image.LoadTrainingGreyImage(dataset_size, './Training/X2_grey/') Y_grey = Image.LoadTrainingGreyImage(dataset_size, './Training/HR_grey/') X_norm = Image.Normalize(X_grey) Y_norm = Image.Normalize(Y_grey) X_cropped = Image.Segment(X_norm, 256) Y_cropped = Image.Segment(Y_norm, 512) # print(X_cropped[0]) X_shuffle, Y_shuffle = Image.Shuffle(X_cropped, Y_cropped) X_final = Image.ExpandDims(X_shuffle) Y_final = Image.ExpandDims(Y_shuffle) Image.SaveOutput([Y_final[0]], './y.png') # print(Y_final[0]) SuperResolution.Test(X_final[0], model_path, './')