operation.py

import time
import numpy as np
import h5py
import matplotlib.pyplot as plt
import scipy.io as sio
import scipy
import os
import math
import h5py
from operator import itemgetter



def load_hscnn():
    imgs = []
    for i in range(150):
        print i
        path = 'training_data/%d.mat' % i
        img = sio.loadmat(path)['data']
        imgs.append(img)
    return imgs

def normalize_0_to_1(mat):
    for i in range(mat.shape[0]):

        mat[i]=(mat[i]-mat[i].min())/(mat[i].max()-mat[i].min())
    return mat


def shuffle_crop(original_data,batch_size):

    index=np.random.choice(range(150),batch_size)

    new_data=[]

    for i in range(batch_size):
        h=original_data[index[i]].shape[0]
        w=original_data[index[i]].shape[1]
        x_index = np.random.randint(0, h - 256)
        y_index = np.random.randint(0, w - 256)
        img=original_data[index[i]][x_index:x_index + 256, y_index:y_index + 256, :]
        new_data.append(img)
    new_data=np.array(new_data)

    return new_data



def load_data_painting(nrf_pairs=10000):
    index = np.random.choice(96000, nrf_pairs)
    train_input=np.zeros((nrf_pairs,256,256,1))
    train_truth=np.zeros((nrf_pairs,256,256,1))
    k=0
    for i in index:
        path = 'training_refine_data/truth_%d.mat' % i
        img = sio.loadmat(path)['data']
        train_truth[k,:,:,0]=img
        path = 'training_refine_data/result_from_unet1_%d.mat' % i
        img = sio.loadmat(path)['data']
        train_input[k,:,:,0]=img
        k+=1
    return train_truth,train_input