Advanced
In addition to basic usage, we also provide advanced settings to give you more flexibility in controlling the output text images. Here, we introduce randomization features, mainly used for training models.
To use the randomization features, you should use the RandomWordCanvas class.
Random Fonts
Use the random_font parameter to enable random font functionality. When random_font is set to True, the font_bank parameter will take effect, and the font_path will be ignored.
You should specify the font_bank parameter pointing to your font library, as the default value is the fonts directory under the package. For example, we have preplaced two fonts in the fonts directory. If you do not modify font_bank, it will randomly choose between these two fonts.
import numpy as np
from wordcanvas import RandomWordCanvas
gen = RandomWordCanvas(
random_font=True,
output_size=(64, 512),
# font_bank="path/to/your/font/bank"
)
imgs = []
for _ in range(8):
text = 'Hello, World!'
img = gen(text)
imgs.append(img)
# Combine all images and output
img = np.concatenate(imgs, axis=0)
Random Text Content
If you're unsure what text to generate, you can use the random_text parameter.
When random_text is set to True, the original text input will be ignored.
import numpy as np
from wordcanvas import RandomWordCanvas
gen = RandomWordCanvas(
random_text=True,
output_size=(64, 512),
)
imgs = []
for _ in range(8):
text = 'Hello!World!' # This input will be ignored
img = gen(text)
imgs.append(img)
# Combine all images and output
img = np.concatenate(imgs, axis=0)
Specify String Length
When random_text is enabled, you can use:
min_random_text_length: minimum text lengthmax_random_text_length: maximum text length
These two parameters allow you to specify the text length range.
import numpy as np
from wordcanvas import RandomWordCanvas
# Generate fixed-length text of 5 characters
gen = RandomWordCanvas(
random_text=True,
min_random_text_length=5,
max_random_text_length=5,
output_size=(64, 512),
)
imgs = []
for _ in range(8):
img = gen()
imgs.append(img)
# Combine all images and output
img = np.concatenate(imgs, axis=0)
Random Background Color
Use the random_background_color parameter to enable random background color functionality.
import numpy as np
from wordcanvas import RandomWordCanvas
gen = RandomWordCanvas(
random_background_color=True,
output_size=(64, 512),
)
imgs = []
for _ in range(8):
text = 'Hello, World!'
img = gen(text)
imgs.append(img)
# Combine all images and output
img = np.concatenate(imgs, axis=0)
Random Text Color
Use the random_text_color parameter to enable the random text color feature.
import numpy as np
from wordcanvas import RandomWordCanvas
gen = RandomWordCanvas(
random_text_color=True,
output_size=(64, 512),
)
imgs = []
for _ in range(8):
text = 'Hello, World!'
img = gen(text)
imgs.append(img)
# Combine all images and output
img = np.concatenate(imgs, axis=0)
Random Text Alignment
Use the random_align_mode parameter to enable the random text alignment feature.
import numpy as np
from wordcanvas import RandomWordCanvas
gen = RandomWordCanvas(
random_align_mode=True,
output_size=(64, 512),
)
imgs = []
for _ in range(8):
text = 'Hello, World!'
img = gen(text)
imgs.append(img)
# Combine all images and output
img = np.concatenate(imgs, axis=0)
Random Text Direction
Use the random_direction parameter to enable the random text direction feature.
It is recommended to use this parameter with output_direction for easier output image handling.
import numpy as np
from wordcanvas import RandomWordCanvas, OutputDirection
gen = RandomWordCanvas(
random_direction=True,
output_direction=OutputDirection.Horizontal,
output_size=(64, 512),
)
imgs = []
for _ in range(8):
text = 'Good afternoon, or good night.'
img = gen(text)
imgs.append(img)
# Combine all images and output
img = np.concatenate(imgs, axis=0)
Random Text Stroke
Use the random_stroke_width parameter to enable the random text stroke feature. You can also use min_random_stroke_width and max_random_stroke_width to specify the stroke width range.
The stroke color can be set via stroke_fill or enable random stroke color using random_stroke_fill.
import numpy as np
from wordcanvas import RandomWordCanvas
gen = RandomWordCanvas(
random_stroke_width=True, # Enable random stroke width
random_stroke_fill=True, # Enable random stroke color
min_random_stroke_width=1, # Minimum stroke width
max_random_stroke_width=5, # Maximum stroke width
output_size=(64, 512),
)
imgs = []
for _ in range(8):
text = 'Hello, World!'
img = gen(text)
imgs.append(img)
# Combine all images and output
img = np.concatenate(imgs, axis=0)
Using random_stroke_width may trigger a warning:
Using `random_stroke_width` may cause an OSError: array allocation size too large error with certain text.
This is a known issue with the `Pillow` library (see https://github.com/python-pillow/Pillow/issues/7287) and cannot be resolved directly.
During testing, we found that using stroke_width in Pillow intermittently causes an OSError. This is a known issue with Pillow, and the related issue link is included in the warning for you to review.
Random Line Spacing
Use the random_spacing parameter to enable the random line spacing feature. You can pair it with min_random_spacing and max_random_spacing to specify the spacing range.
import numpy as np
from wordcanvas import RandomWordCanvas, AlignMode
gen = RandomWordCanvas(
random_spacing=True,
min_random_spacing=10,
max_random_spacing=100,
align_mode=AlignMode.Center,
output_size=(128, 512),
)
imgs = []
for _ in range(2):
img = gen('你好！\nHello, World!')
imgs.append(img)
# Combine all images and output
img = np.concatenate(imgs, axis=1)
Font Weight
Due to variations in the number of characters supported by each font, we might encounter uneven font weights during model training. To alleviate this, you can use the random_font_weight parameter.
In simple terms, since each font has an equal chance of being selected, but some characters are only supported by a few fonts, you may notice that certain characters are rarely trained on.
from wordcanvas import RandomWordCanvas
gen = RandomWordCanvas(
random_font=True,
random_font_weight=True,
output_size=(64, 512),
)
When this parameter is enabled, RandomWordCanvas adjusts the font selection probability based on the number of characters a font supports. Fonts with fewer supported characters are less likely to be selected, achieving a more balanced distribution.
Block List
While using fonts, we have found that some font files are inconsistent.
For example, while the list of characters supported by a font can be read from the font file, certain characters may fail to render correctly during actual use. To address this, we have developed a block list feature that allows you to exclude problematic fonts.
Use the block_font_list parameter to set the block list:
import numpy as np
from wordcanvas import WordCanvas, RandomWordCanvas
gen = WordCanvas(
block_font_list=['AABB']
)
gen_random = RandomWordCanvas(
block_font_list=['AABB'],
)
After setting this, the program will automatically filter out fonts containing *AABB*.(ttf|otf) and prevent them from being selected.
Dashboard
Returning to the dashboard, the information for RandomWordCanvas is even more detailed:
When randomization parameters are enabled, parameters set to True are marked in green, while those set to False are marked in red.
This design aims to help you quickly confirm the relevant settings.
Summary
In the development of these tools, our goal was to create a versatile tool capable of generating various text images, especially for deep learning model training.
The introduction of randomness aims to simulate various real-world scenarios, which is highly beneficial for improving model adaptability and generalization. We hope you find these features useful!