In [3]:

    

import numpy as np
import os
from scipy.misc import imread, imresize
import matplotlib.pyplot as plt
%matplotlib inline
print ("Packages are loaded!!!")
cwd = os.getcwd()
print ("Current folder is %s" % (cwd))


    

    




Packages are loaded!!!
Current folder is /home/parksurk/dl/dl_with_tf

In [7]:

    

# Training set folder
paths = {"./images/celebs/Adam_Levine"
        , "./images/celebs/Jennifer_Lawrence"
        , "./images/celebs/Barack_Obama"
        , "./images/celebs/Donal_Trump"}
# The reshape size
imgsize = [64, 64]
# Grayscale
use_gray = 1
# Save name
data_name = "custom_data"

print ("Your images should be at")
for i, path in enumerate(paths):
    print (" [%d/%d] %s/%s" % (i, len(paths), cwd, path))
    
print ("Data will be saved to %s" % (cwd + '/data/' + data_name + '.npz'))


    

    




Your images should be at
 [0/4] /home/parksurk/dl/dl_with_tf/./images/celebs/Jennifer_Lawrence
 [1/4] /home/parksurk/dl/dl_with_tf/./images/celebs/Adam_Levine
 [2/4] /home/parksurk/dl/dl_with_tf/./images/celebs/Donal_Trump
 [3/4] /home/parksurk/dl/dl_with_tf/./images/celebs/Barack_Obama
Data will be saved to /home/parksurk/dl/dl_with_tf/data/custom_data.npz

In [8]:

    

def rgb2gray(rgb):
    if len(rgb.shape) is 3:
        return np.dot(rgb[...,:3], [0.299, 0.587, 0.114])
    else:
        # print ("Current Image if GRAY!")
        return rgb


    

In [9]:

    

nclass = len(paths)
valid_exts = {".jpg", ".gif", ".png", ".tga", ".jpeg"}
imgcnt = 0
for i, relpath in zip(range(nclass), paths):
    path = cwd + "/" + relpath
    flist = os.listdir(path)
    for f in flist:
        if os.path.splitext(f)[1].lower() not in valid_exts:
            continue
        fullpath = os.path.join(path, f)
        currimg = imread(fullpath)
        # Convert to grayscale
        if use_gray:
            grayimg = rgb2gray(currimg)
        else:
            grayimg = currimg
        # Reshape
        graysmall = imresize(grayimg, [imgsize[0], imgsize[1]])/255.
        grayvec = np.reshape(graysmall, (1, -1))
        # Save
        curr_label =np.eye(nclass, nclass)[i:i+1, :]
        if imgcnt is 0:
            totalimg = grayvec
            totallabel = curr_label
        else:
            totalimg = np.concatenate((totalimg, grayvec), axis=0)
            totallabel = np.concatenate((totallabel, curr_label), axis=0)
        imgcnt = imgcnt + 1
print ("Total %d images loaded." % (imgcnt))


    

    




Total 404 images loaded.

In [11]:

    

def print_shape(string, x):
    print ("Shape of '%s' is %s" % (string, x.shape,))
    
randidx = np.random.randint(imgcnt, size=imgcnt)
trainidx = randidx[0:int(3*imgcnt/5)]
testidx = randidx[int(3*imgcnt/5):imgcnt]
trainimg = totalimg[trainidx, :]
trainlabel = totallabel[trainidx, :]
testimg = totalimg[testidx, :]
testlabel = totallabel[testidx, :]
print_shape("trainimg", trainimg)
print_shape("trainlabel", trainlabel)
print_shape("testimg", testimg)
print_shape("testlabel", testlabel)


    

    




Shape of 'trainimg' is (242, 4096)
Shape of 'trainlabel' is (242, 4)
Shape of 'testimg' is (162, 4096)
Shape of 'testlabel' is (162, 4)

In [14]:

    

savepath = cwd + "/data/" + data_name + ".npz"
np.savez(savepath, trainimg=trainimg, trainlabel=trainlabel, testimg=testimg, testlabel=testlabel
        , imgsize=imgsize, use_gray=use_gray)
print("Saved to %s" % (savepath))


    

    




Saved to /home/parksurk/dl/dl_with_tf/data/custom_data.npz

In [15]:

    

# Load images from npz file
cwd = os.getcwd()
loadpath = cwd + "/data/" + data_name + ".npz"
l = np.load(loadpath)

# See what's in 
l.files

# Parse data
trainimg_loaded = l['trainimg']
trainlabel_loaded = l['trainlabel']
testimg_loaded = l['testimg']
testlabel_loaded = l['testlabel']

print ("%d train images loaded" % (trainimg_loaded.shape[0]))
print ("%d test images loaded" % (testimg_loaded.shape[0]))
print ("Loaded from to %s" % (savepath))


    

    




242 train images loaded
162 test images loaded
Loaded from to /home/parksurk/dl/dl_with_tf/data/custom_data.npz

In [16]:

    

ntrain_loaded = trainimg_loaded.shape[0]
batch_size = 10;
randidx = np.random.randint(ntrain_loaded, size=batch_size)
for i in randidx:
    #currimg = np.reshape(trainimg_loaded[i, :], (imgsize[0], -1))
    currlabel_onehot = trainlabel_loaded[i, :]
    currlabel = np.argmax(currlabel_onehot)
    if use_gray:
        currimg = np.reshape(trainimg_loaded[i, :], (imgsize[0], -1))
        plt.matshow(currimg, cmap=plt.get_cmap('gray'))
        plt.colorbar()
    else:
        currimg = np.reshape(trainimg_loaded[i, :], (imgsize[0], imgsize[1], 3))
        plt.imshow(currimg)
    title_string = "[%d] %d-class" % (i, currlabel)
    plt.title(title_string)
    plt.show()


    

In [17]:

    

# Do batch stuff using loaded data
ntest_loaded = testimg_loaded.shape[0]
batch_size = 3
randidx = np.random.randint(ntest_loaded, size=batch_size)
for i in randidx:
    #currimg = np.reshape(testimg_loaded[i, :], (imgsize[0], -1))
    currlabel_onehot = testimg_loaded[i, :]
    currlabel = np.argmax(currlabel_onehot)
    if use_gray:
        currimg = np.reshape(testimg_loaded[i, :], (imgsize[0], -1))
        plt.matshow(currimg, cmap=plt.get_cmap('gray'))
        plt.colorbar()
    else:
        currimg = np.reshape(testimg_loaded[i, :], (imgsize[0], imgsize[1], 3))
        plt.imshow(currimg)
    title_string = "[%d] %d-class" % (i, currlabel)
    plt.title(title_string)
    plt.show()


    

In [ ]:

    




    

In [ ]:

    




    

In [ ]:

    




    

In [ ]:

    




    

In [ ]: