def main():
parser = argparse.ArgumentParser(
description='Converts the IAM-Offline dataset to a graves version.')
parser.add_argument('input', help='The input dataset')
parser.add_argument('output', help='The output dataset')
parser.add_argument('--output-skeletons',
required=True,
help='An output folder for skeletonization results')
args = parser.parse_args()
print(args)
iamOfflineDataset = IamOfflineDataset(args.input)
if args.output_skeletons and not os.path.isdir(args.output_skeletons):
os.makedirs(args.output_skeletons)
skeletonDataset = dict()
with Skeletonizer() as skeletonizer:
for formId, formImg in tqdm(iamOfflineDataset.formsIterator()):
outFile = os.path.join(args.output_skeletons, formId + '.png')
outSkelFile = os.path.join(args.output_skeletons,
formId + '_skel.png')
skeletonBlurImg = None
skeletonImg = None
# Use cacheing if skeleton dir is provided
if os.path.isfile(outFile):
skeletonBlurImg = Image.open(outFile)
if os.path.isfile(outSkelFile):
skeletonImg = Image.open(outSkelFile)
print('Found cached file: ' + outSkelFile)
if not skeletonBlurImg:
skeletonBlurImg = skeletonizer.skeletonize_blurred(formImg)
skeletonBlurImg.save(outFile)
if not skeletonImg:
skeletonImgRaw = skeletonizer.skeletonize_sharp(
skeletonBlurImg)
skeletonImg = Image.fromarray(
skeletonImgRaw.astype('uint8') * 255)
skeletonImg.save(outSkelFile)
skeletonDataset[formId] = outSkelFile
outputDataset = ConvertedGravesDataset()
def imgLoader(form_id):
return Image.open(skeletonDataset[form_id])
for (lineId, lineImg,
lineText) in tqdm(iamOfflineDataset.linesIterator(imgLoader)):
# print('SkeletonImg:', lineId, lineImg, lineText)
penPositions = sample_to_penpositions(lineImg)
outputDataset.addSample(penPositions, lineText)
outputDataset.save(args.output)
def main():
parser = argparse.ArgumentParser(
description=
'The first working version (without pen style transfer). Modifies the content of an image.'
)
parser.add_argument('--text-in', help='The input text', required=True)
parser.add_argument('--text-out', help='The output text', required=True)
parser.add_argument('input', help='The input file')
args = parser.parse_args()
print(args)
inputImg = Image.open(args.input)
with Skeletonizer() as skeletonizer:
skeletonBlurImg = skeletonizer.skeletonize_blurred(inputImg)
skeletonImg = skeletonizer.skeletonize_sharp(skeletonBlurImg)
penPositions = sample_to_penpositions(skeletonImg)
with GravesWriter() as writer:
newPenPositions = writer.write(args.text_out, args.text_in,
penPositions)
newPenPositions = align(newPenPositions, penPositions)
newSkeletonBlurImg, newSkeletonImg = render_skeleton(
newPenPositions, inputImg.size)
with PenStyleTransfer() as penStyleTransfer:
outputImg = penStyleTransfer.transferStyle(newSkeletonBlurImg,
inputImg)
print("Done. Displaying results ...")
plt.figure('Full Pipeline', figsize=(16, 9))
plt.subplot(3, 2, 1)
plt.imshow(inputImg)
plt.subplot(3, 2, 3)
plt.imshow(skeletonBlurImg)
plt.subplot(3, 2, 5)
plt.imshow(skeletonImg, cmap='binary', vmax=10)
plotPenPositions(penPositions)
plt.subplot(3, 2, 6)
plt.imshow(newSkeletonImg, cmap='binary', vmax=256 * 10)
plotPenPositions(newPenPositions)
plt.subplot(3, 2, 4)
plt.imshow(newSkeletonBlurImg)
plt.subplot(3, 2, 2)
plt.imshow(outputImg)
plt.show()
def main():
parser = argparse.ArgumentParser(
description='Converts a dataset of skeletons to a graves dataset.')
parser.add_argument('input', help='The input skeletons folder')
parser.add_argument('output', help='The output folder')
parser.add_argument(
'--sort-by',
default='mean',
help=
'The sorting order. Can be one of \'first\', \'last\', \'leftmost\', \'rightmost\', \'mean\'.'
)
args = parser.parse_args()
print(args)
fileNames = [
f for f in os.listdir(args.input)
if os.path.isfile(os.path.join(args.input, f))
]
if not os.path.isdir(args.output):
os.makedirs(args.output)
for fileName in tqdm(fileNames):
#print(fileName)
strokeId = os.path.splitext(os.path.basename(fileName))[0]
# print(strokeId, lineText)
fullName = os.path.join(args.input, fileName)
lineImg = ImageOps.invert(Image.open(fullName).convert('L'))
# print('SkeletonImg:', lineId, lineImg, lineText)
penPositions = sample_to_penpositions(lineImg, args.sort_by)
svg = svgwrite.Drawing(os.path.join(args.output, strokeId + '.svg'),
profile='tiny',
size=(str(lineImg.width) + 'px',
str(lineImg.height) + 'px'))
svg.add(svg.rect(size=('100%', '100%'), fill='white'))
render_penpositions_to_svg(penPositions, svg)
svg.save()
def main():
parser = argparse.ArgumentParser(
description='Converts a dataset of skeletons to a graves dataset.')
parser.add_argument('input', help='The input skeletons folder')
parser.add_argument('labels', help='The text labels as .npy file')
parser.add_argument('output', help='The output dataset')
parser.add_argument(
'--sort-by',
default='mean',
help=
'The sorting order. Can be one of \'first\', \'last\', \'leftmost\', \'rightmost\', \'mean\'.'
)
args = parser.parse_args()
print(args)
fileNames = [
f for f in os.listdir(args.input)
if os.path.isfile(os.path.join(args.input, f))
]
textAnnotations = np.load(args.labels, allow_pickle=True).item()
outputDataset = ConvertedGravesDataset()
for fileName in tqdm(fileNames):
#print(fileName)
strokeId = os.path.splitext(os.path.basename(fileName))[0]
lineText = textAnnotations[strokeId]
# print(strokeId, lineText)
fullName = os.path.join(args.input, fileName)
lineImg = ImageOps.invert(Image.open(fullName).convert('L'))
# print('SkeletonImg:', lineId, lineImg, lineText)
penPositions = sample_to_penpositions(lineImg, args.sort_by)
outputDataset.addSample(penPositions, lineText)
outputDataset.save(args.output)
def main():
parser = argparse.ArgumentParser(description='The first working version (without pen style transfer). Modifies the content of an image.')
parser.add_argument('--text-in', help='The input text', required=True)
parser.add_argument('--text-out', help='The output text', required=True)
parser.add_argument('input', help='The input file')
parser.add_argument('--output', help='The output file')
args = parser.parse_args()
print(args)
inputImg = Image.open(args.input)
with ForegroundExtractor() as foregroundExtractor:
foreground = foregroundExtractor.extract_foreground(inputImg)
backgroundMask = foregroundExtractor.create_background_mask(foreground)
#print(np.asarray(foreground) / 255.0)
#print(np.asarray(backgroundMask))
background_masked = (np.asarray(inputImg)/255.0) * backgroundMask
background_img = Image.fromarray(np.maximum(0, np.minimum(255, background_masked * 255)).astype(np.uint8))
background_img.save('tmp.png')
foreground.save('tmp2.png')
with BackgroundFiller() as backgroundFiller:
background = backgroundFiller.fill(inputImg, backgroundMask)
background.save('tmp3.png')
with Skeletonizer() as skeletonizer:
skeletonBlurImg = skeletonizer.skeletonize_blurred(inputImg)
skeletonImg = skeletonizer.skeletonize_sharp(skeletonBlurImg)
penPositions = sample_to_penpositions(skeletonImg)
with GravesWriter() as writer:
newPenPositions = writer.write(args.text_out, args.text_in, penPositions)
newPenPositions = align(newPenPositions, penPositions)
newSkeletonBlurImg, newSkeletonImg = render_skeleton(newPenPositions, inputImg.size)
#with Colorizer() as colorizer:
# outputImg = colorizer.colorize(newSkeletonBlurImg)
with PenStyleTransfer() as penStyleTransfer:
outputImg = penStyleTransfer.transferStyle(newSkeletonBlurImg, foreground)
blendedImg = blend(outputImg, background)
print("Done. Displaying results ...")
plt.figure('Full Pipeline', figsize=(16, 9))
plt.subplot(3, 3, 1)
plt.imshow(inputImg)
plt.subplot(3, 3, 2)
plt.imshow(foreground)
plt.subplot(3, 3, 5)
plt.imshow(background_masked)
plt.subplot(3, 3, 8)
plt.imshow(background)
plt.subplot(3, 3, 4)
plt.imshow(skeletonBlurImg)
plt.subplot(3, 3, 7)
plt.imshow(skeletonImg, cmap='binary', vmax=10)
plotPenPositions(penPositions)
plt.subplot(3, 3, 9)
plt.imshow(newSkeletonImg, cmap='binary', vmax=256*10)
plotPenPositions(newPenPositions)
plt.subplot(3, 3, 6)
plt.imshow(newSkeletonBlurImg)
plt.subplot(3, 3, 3)
plt.imshow(blendedImg)
plt.show()
if args.output:
Image.fromarray(255 - skeletonImg.astype(np.uint8)*255).save(args.output)