A library to generate pictures of text for the purposes of training machine learning models. Inspired by this Dropbox blog post on building an OCR pipeline.
python generate_text_pics.py [-cwspa] CORPUS.txt FONTS_DIR [-rsbucp]
Optional parameters at the beginning of the command specify which tokens you would like to create (default is -w):
-c: create images of the characters in the corpus
-w: create images of the words in the corpus
-s: create images of the sentences in the corpus
-p: create images of the paragraphs in the corpus
-a: create images of the entire corpus
Optional parameters at the end of the command create images in addition to an unmodified base image for each word in each font:
-r: create images that are rotated slightly CW and CCW
-s: create images that are skewed slightly left and right
-b: create images with a slight gaussian blur
-u: create images with an underline
-c: create images with a complex transformation composed of skewing and blurring
-p: removes all punctuation from words
python src/generate_text_pics.py -wsp sample_text/lorem_ipsum.txt fonts/some/ -sbc
- modularize argument parsing
- make argument parsing more robust