def images_to_csv_with_label(csv_file, images, label, width, height):
csv_file.write("label," + ",".join(["pixel" + str(i) for i in range(0, int(width) * int(height))]) + "\n")
for i, nameFile in enumerate(images):
print(nameFile)
csv_file.write(
label + "," + ",".join(
str(x) for x in misc.img_to_1d_gray(Image(nameFile).resize(width, height))) + "\n")
def images_to_csv_with_label(csv_file, images, label, width, height):
csv_file.write("label," + ",".join(
["pixel" + str(i) for i in range(0,
int(width) * int(height))]) + "\n")
for i, nameFile in enumerate(images):
print(nameFile)
csv_file.write(label + "," + ",".join(
str(x) for x in misc.img_to_1d_gray(
Image(nameFile).resize(width, height))) + "\n")
def segmented_to_csv(path="", csv_path="train.csv", folder_to_label_dict={}, width=28, height=28):
with open(csv_path, "w") as train_csv:
train_csv.write("label," + ",".join(["pixel" + str(i) for i in range(0, int(width) * int(height))]) + "\n")
# print(folder_to_label_dict)
for folder in folder_to_label_dict.keys():
# print(folder)
files = glob.glob(path + folder + "/*")
for img_name in files:
# print(img_name)
train_csv.write(folder_to_label_dict[folder] + "," + ",".join(
str(x) for x in misc.img_to_1d_gray(Image(img_name).resize(width, height))) + "\n")
def create_train_csv(path=None, width=28, height=28):
"""create a train csv file from image files and their labels"""
if path is None:
path = Popen("pwd", shell=True, stdin=PIPE, stdout=PIPE).stdout.read()[:-1]
with open(path + "/train.csv", "w") as train_csv, open(
path + "/trainLabels.csv", "r") as label_maps_csv:
train_csv.write("label," + ",".join(["pixel" + str(i) for i in range(0, int(width) * int(height))]) + "\n")
files = glob.glob(path + "/trainResized/*")
labels = {line.strip().split(",")[0]: line.strip().split(",")[1] for line in label_maps_csv}
for i, nameFile in enumerate(files):
file_id = strip_extension(get_file_name(nameFile))
train_csv.write(labels[file_id] + "," + ",".join(str(x) for x in misc.img_to_1d_gray(nameFile)) + "\n")
def create_test_csv(path=None, width=28, height=28):
"""create a test csv file and their id list from image files"""
if path is None:
path = Popen("pwd", shell=True, stdin=PIPE, stdout=PIPE).stdout.read()[:-1]
with open(path + "/test.csv", "w") as test_csv, open(
path + "/testIDs.csv", "w") as test_ids_csv:
test_csv.write(",".join(["pixel" + str(i) for i in range(0, int(width) * int(height))]) + "\n")
test_ids_csv.write("ID," + "\n")
files = glob.glob(path + "/testResized/*")
for i, nameFile in enumerate(files):
file_id = strip_extension(get_file_name(nameFile))
test_ids_csv.write(file_id + "\n") # save the order of test data (which line is which file)
test_csv.write(",".join(str(x) for x in misc.img_to_1d_gray(nameFile)) + "\n")
def segmented_to_csv(path="",
csv_path="train.csv",
folder_to_label_dict={},
width=28,
height=28):
with open(csv_path, "w") as train_csv:
train_csv.write("label," + ",".join(
["pixel" + str(i)
for i in range(0,
int(width) * int(height))]) + "\n")
# print(folder_to_label_dict)
for folder in folder_to_label_dict.keys():
# print(folder)
files = glob.glob(path + folder + "/*")
for img_name in files:
# print(img_name)
train_csv.write(folder_to_label_dict[folder] + "," + ",".join(
str(x) for x in misc.img_to_1d_gray(
Image(img_name).resize(width, height))) + "\n")
def create_train_csv(path=None, width=28, height=28):
"""create a train csv file from image files and their labels"""
if path is None:
path = Popen("pwd", shell=True, stdin=PIPE,
stdout=PIPE).stdout.read()[:-1]
with open(path + "/train.csv",
"w") as train_csv, open(path + "/trainLabels.csv",
"r") as label_maps_csv:
train_csv.write("label," + ",".join(
["pixel" + str(i)
for i in range(0,
int(width) * int(height))]) + "\n")
files = glob.glob(path + "/trainResized/*")
labels = {
line.strip().split(",")[0]: line.strip().split(",")[1]
for line in label_maps_csv
}
for i, nameFile in enumerate(files):
file_id = strip_extension(get_file_name(nameFile))
train_csv.write(labels[file_id] + "," + ",".join(
str(x) for x in misc.img_to_1d_gray(nameFile)) + "\n")
def create_test_csv(path=None, width=28, height=28):
"""create a test csv file and their id list from image files"""
if path is None:
path = Popen("pwd", shell=True, stdin=PIPE,
stdout=PIPE).stdout.read()[:-1]
with open(path + "/test.csv",
"w") as test_csv, open(path + "/testIDs.csv",
"w") as test_ids_csv:
test_csv.write(",".join(
["pixel" + str(i)
for i in range(0,
int(width) * int(height))]) + "\n")
test_ids_csv.write("ID," + "\n")
files = glob.glob(path + "/testResized/*")
for i, nameFile in enumerate(files):
file_id = strip_extension(get_file_name(nameFile))
test_ids_csv.write(
file_id +
"\n") # save the order of test data (which line is which file)
test_csv.write(",".join(
str(x) for x in misc.img_to_1d_gray(nameFile)) + "\n")
def generate_letter_images():
"""generate a new data of warped characters [a-zA-Z0-9]"""
if "(" in sys.argv[1]: # assume a dimension input and an empty bg
[w, h] = sys.argv[1][1:-1:].split(',')
print(int(w) * int(h))
with open("/home/agp/workspace/deep_learning/streetView/new_dataset/train_with_inverted.csv", "w") as train_csv:
train_csv.write("label," + ",".join(["pixel" + str(i) for i in range(0, int(w) * int(h))]) + "\n")
for text in string.ascii_letters + string.digits: # for all alphanumeric
print(text)
if "(" in sys.argv[1]: # assume a dimension input and an empty bg
img = Image(tuple([int(i) for i in sys.argv[1][1:-1:].split(',')]))
fname = "test_input.Bmp"
else:
img = Image(sys.argv[1]).invert() # background e.g. Image((150,100))#
fname = sys.argv[1]
path = "/home/agp/workspace/deep_learning/streetView/new_dataset/"
sys.stdout.flush()
# text = sys.argv[2] # e.g. "A.G.Polat"
w = img.width
h = img.height
# extended fonts: wont work in windows!
if False and "Linux" in platform.system(): # skip them for now
fonts = Popen('find /usr/share/fonts | grep ttf', shell=True, stdin=PIPE,
stdout=PIPE).stdout.read().strip().split('\n') + img.dl().listFonts()
else:
fonts = img.dl().listFonts()
# print(fonts)
# fonts=xfonts+fonts
colors = [(random.randint(0, 255), random.randint(0, 255), random.randint(0, 255)) for _ in
range(0, len(fonts))]
i = 0
newImg = img.copy()
img2 = Image(img.size())
for font in fonts:
if font == "" or font in ["qbicle1brk", "dblayer1brk", "dblayer3brk", "droidsanshebrew", "mrykacstqurn",
"zoetropebrk", "lohitpunjabi", "loopybrk", "kacstqurn", "droidsansgeorgian",
"mallige", "webdings", "3dletbrk", "scalelinesmazebrk", "kacstpen",
"nucleusbrk", "unresponsivebrk", "kacstart", "kacstdecorative", "dblayer2brk",
"kacstoffice", "binaryxbrk", "binaryx01sbrk", "lklug", "symmetrybrk",
"linedingsbrk", "droidsansarmenian", "zurklezoutlinebrk",
"entangledlayerbbrk", "kacstdigital", "skullcapzbrk", "kacstbook",
"konectoro1brk", "yourcomplexobrk", "opensymbol", "18holesbrk",
"droidsansjapanese", "qbicle4brk",
"qbicle3brk", "lohitdevanagari", "doublebogeybrk", "binary01sbrk",
"kacstscreen", "lohitbengali", "kacstposter", "kedage", "kacstone",
"kacstnaskh", "droidsansethiopic", "qbicle2brk", "vemana2000", "kacstfarsi",
"xmaslightsbrk", "headdingmakerbrk", "kacstletter", "binarybrk", "ori1uni",
"90starsbrk", "codeoflifebrk", "saab", "zurklezsolidbrk", "pothana2000",
"kacsttitle", "fauxsnowbrk", "lohitgujarati", "tetricidebrk", "lohittamil",
"kacsttitlel", "droidsansthai", "dblayer4brk", "bitblocksttfbrk",
"pindownbrk", "droidarabicnaskh"]: # font is no good
continue
# if text == "":
# text=font.split("/")[-1]
img2.dl().selectFont(font)
y = random.randint(0, h)
x = random.randint(0, w)
fs = int(w / 1.1)
fontSize = random.randint(random.randint(fs, w), int(w * 1.1))
img2.dl().setFontSize(fontSize) # font size between 2 and 100
(w1, h1) = img2.dl().textDimensions(text)
k = 0
skip_this_one = False
while x + w1 > w * 1.1 or y + h1 > h * 1.1: # 1.2 for allowing text to be partly outside the box
k += 1
y = random.randint(0, h)
x = random.randint(0, w)
# if k>5:
# y=random.randint(0,h//3+int(random.random()*0.6*h))
# x=random.randint(0,int(w*random.random()*0.4))
if k > 500:
skip_this_one = True # the font is too big for the image dimensions
print("skipping the big font:" + str(font))
break
fontSize = int(0.95 * fontSize)
if fontSize < w // 5:
fontSize = random.randint(random.randint(fs, w - 1), w)
fs -= 2
if fs < 3:
fs = random.randint(2, w - 1)
img2.dl().setFontSize(fontSize)
(w1, h1) = img2.dl().textDimensions(text)
if skip_this_one:
img2.clearLayers()
newImg.clearLayers()
continue
img2.drawText(text, x, y, colors[i], fontSize)
if random.randint(0, 10) < 2:
img = newImg + img2.applyLayers()
train_csv.write(text + "," + ",".join(str(x) for x in misc.img_to_1d_gray(img)) + "\n")
if random.randint(0, 10) < 1:
img = newImg + img2.applyLayers().warp(clockwise_quadrilateral(w, h))
train_csv.write(text + "," + ",".join(str(x) for x in misc.img_to_1d_gray(img)) + "\n")
if random.randint(0, 10) < 1:
img = newImg + img2.applyLayers().rotate(random.randint(-90, 90))
train_csv.write(text + "," + ",".join(str(x) for x in misc.img_to_1d_gray(img)) + "\n")
if random.randint(0, 10) < 8: # warp
img = newImg + img2.applyLayers().rotate(random.randint(-90, 90)).warp(
clockwise_quadrilateral(w, h))
train_csv.write(text + "," + ",".join(str(x) for x in misc.img_to_1d_gray(img)) + "\n")
if random.randint(0, 10) < 5: # salt&pepper
img = add_noise(img, random.random() * 0.0025)
train_csv.write(text + "," + ",".join(str(x) for x in misc.img_to_1d_gray(img)) + "\n")
# img=img.smooth(sigma=random.random()*3,spatial_sigma=random.random()*3)
# if random.randint(0,10)<1:
# img=add_noise(img,random.random()*0.001)
img2.clearLayers()
if random.randint(0, 10) < 11: # always true
img = img.blur((random.randint(1, random.randint(2, fs // 5 + 4)),
random.randint(1, random.randint(2, fs // 5 + 4))))
if random.randint(0, 10) < 7:
img = img.invert()
train_csv.write(text + "," + ",".join(str(x) for x in misc.img_to_1d_gray(img)) + "\n")
# img.show()
# time.sleep(0.5)
# file_to_save = fname[:-4:] + "_" + text + str(font.replace("/", "_")) + fname[-4::]
# print(file_to_save)
# img.save(path + file_to_save) # str(font.replace("/","_"))+fname[-4::])
# Popen(['/home/agp/Desktop/scenetext/code/DetectText/./DetectText',
# file_to_save,
# '/home/agp/Desktop/scenetext/code/DetectText/result.png', '1'])
# Popen('rm temp.png',shell=True,stdin=PIPE,stdout=PIPE)
# print("swt:")
# Image('SWT.png').show()
# time.sleep(5)
# else:
# img.save(sys.argv[1][:-4:]+"_"+str(font)+sys.argv[1][-4::])
# newImg.clearLayers()
i += 1
def generate_letter_images():
"""generate a new data of warped characters [a-zA-Z0-9]"""
if "(" in sys.argv[1]: # assume a dimension input and an empty bg
[w, h] = sys.argv[1][1:-1:].split(',')
print(int(w) * int(h))
with open(
"/home/agp/workspace/deep_learning/streetView/new_dataset/train_with_inverted.csv",
"w") as train_csv:
train_csv.write(
"label," +
",".join(["pixel" + str(i)
for i in range(0,
int(w) * int(h))]) + "\n")
for text in string.ascii_letters + string.digits: # for all alphanumeric
print(text)
if "(" in sys.argv[1]: # assume a dimension input and an empty bg
img = Image(
tuple([int(i) for i in sys.argv[1][1:-1:].split(',')]))
fname = "test_input.Bmp"
else:
img = Image(
sys.argv[1]).invert() # background e.g. Image((150,100))#
fname = sys.argv[1]
path = "/home/agp/workspace/deep_learning/streetView/new_dataset/"
sys.stdout.flush()
# text = sys.argv[2] # e.g. "A.G.Polat"
w = img.width
h = img.height
# extended fonts: wont work in windows!
if False and "Linux" in platform.system(): # skip them for now
fonts = Popen('find /usr/share/fonts | grep ttf',
shell=True,
stdin=PIPE,
stdout=PIPE).stdout.read().strip().split(
'\n') + img.dl().listFonts()
else:
fonts = img.dl().listFonts()
# print(fonts)
# fonts=xfonts+fonts
colors = [(random.randint(0, 255), random.randint(0, 255),
random.randint(0, 255)) for _ in range(0, len(fonts))]
i = 0
newImg = img.copy()
img2 = Image(img.size())
for font in fonts:
if font == "" or font in [
"qbicle1brk", "dblayer1brk", "dblayer3brk",
"droidsanshebrew", "mrykacstqurn", "zoetropebrk",
"lohitpunjabi", "loopybrk", "kacstqurn",
"droidsansgeorgian", "mallige", "webdings", "3dletbrk",
"scalelinesmazebrk", "kacstpen", "nucleusbrk",
"unresponsivebrk", "kacstart", "kacstdecorative",
"dblayer2brk", "kacstoffice", "binaryxbrk",
"binaryx01sbrk", "lklug", "symmetrybrk",
"linedingsbrk", "droidsansarmenian",
"zurklezoutlinebrk", "entangledlayerbbrk",
"kacstdigital", "skullcapzbrk", "kacstbook",
"konectoro1brk", "yourcomplexobrk", "opensymbol",
"18holesbrk", "droidsansjapanese", "qbicle4brk",
"qbicle3brk", "lohitdevanagari", "doublebogeybrk",
"binary01sbrk", "kacstscreen", "lohitbengali",
"kacstposter", "kedage", "kacstone", "kacstnaskh",
"droidsansethiopic", "qbicle2brk", "vemana2000",
"kacstfarsi", "xmaslightsbrk", "headdingmakerbrk",
"kacstletter", "binarybrk", "ori1uni", "90starsbrk",
"codeoflifebrk", "saab", "zurklezsolidbrk",
"pothana2000", "kacsttitle", "fauxsnowbrk",
"lohitgujarati", "tetricidebrk", "lohittamil",
"kacsttitlel", "droidsansthai", "dblayer4brk",
"bitblocksttfbrk", "pindownbrk", "droidarabicnaskh"
]: # font is no good
continue
# if text == "":
# text=font.split("/")[-1]
img2.dl().selectFont(font)
y = random.randint(0, h)
x = random.randint(0, w)
fs = int(w / 1.1)
fontSize = random.randint(random.randint(fs, w), int(w * 1.1))
img2.dl().setFontSize(fontSize) # font size between 2 and 100
(w1, h1) = img2.dl().textDimensions(text)
k = 0
skip_this_one = False
while x + w1 > w * 1.1 or y + h1 > h * 1.1: # 1.2 for allowing text to be partly outside the box
k += 1
y = random.randint(0, h)
x = random.randint(0, w)
# if k>5:
# y=random.randint(0,h//3+int(random.random()*0.6*h))
# x=random.randint(0,int(w*random.random()*0.4))
if k > 500:
skip_this_one = True # the font is too big for the image dimensions
print("skipping the big font:" + str(font))
break
fontSize = int(0.95 * fontSize)
if fontSize < w // 5:
fontSize = random.randint(random.randint(fs, w - 1), w)
fs -= 2
if fs < 3:
fs = random.randint(2, w - 1)
img2.dl().setFontSize(fontSize)
(w1, h1) = img2.dl().textDimensions(text)
if skip_this_one:
img2.clearLayers()
newImg.clearLayers()
continue
img2.drawText(text, x, y, colors[i], fontSize)
if random.randint(0, 10) < 2:
img = newImg + img2.applyLayers()
train_csv.write(text + "," + ",".join(
str(x) for x in misc.img_to_1d_gray(img)) + "\n")
if random.randint(0, 10) < 1:
img = newImg + img2.applyLayers().warp(
clockwise_quadrilateral(w, h))
train_csv.write(text + "," + ",".join(
str(x) for x in misc.img_to_1d_gray(img)) + "\n")
if random.randint(0, 10) < 1:
img = newImg + img2.applyLayers().rotate(
random.randint(-90, 90))
train_csv.write(text + "," + ",".join(
str(x) for x in misc.img_to_1d_gray(img)) + "\n")
if random.randint(0, 10) < 8: # warp
img = newImg + img2.applyLayers().rotate(
random.randint(-90, 90)).warp(
clockwise_quadrilateral(w, h))
train_csv.write(text + "," + ",".join(
str(x) for x in misc.img_to_1d_gray(img)) + "\n")
if random.randint(0, 10) < 5: # salt&pepper
img = add_noise(img, random.random() * 0.0025)
train_csv.write(text + "," + ",".join(
str(x) for x in misc.img_to_1d_gray(img)) + "\n")
# img=img.smooth(sigma=random.random()*3,spatial_sigma=random.random()*3)
# if random.randint(0,10)<1:
# img=add_noise(img,random.random()*0.001)
img2.clearLayers()
if random.randint(0, 10) < 11: # always true
img = img.blur(
(random.randint(1, random.randint(2, fs // 5 + 4)),
random.randint(1, random.randint(2, fs // 5 + 4))))
if random.randint(0, 10) < 7:
img = img.invert()
train_csv.write(text + "," + ",".join(
str(x) for x in misc.img_to_1d_gray(img)) + "\n")
# img.show()
# time.sleep(0.5)
# file_to_save = fname[:-4:] + "_" + text + str(font.replace("/", "_")) + fname[-4::]
# print(file_to_save)
# img.save(path + file_to_save) # str(font.replace("/","_"))+fname[-4::])
# Popen(['/home/agp/Desktop/scenetext/code/DetectText/./DetectText',
# file_to_save,
# '/home/agp/Desktop/scenetext/code/DetectText/result.png', '1'])
# Popen('rm temp.png',shell=True,stdin=PIPE,stdout=PIPE)
# print("swt:")
# Image('SWT.png').show()
# time.sleep(5)
# else:
# img.save(sys.argv[1][:-4:]+"_"+str(font)+sys.argv[1][-4::])
# newImg.clearLayers()
i += 1
