Image/PDF synthesizer for Cradl

Disclaimer

This code does not guarantee that images/PDFs will be successfully synthesized. Use at your own risk.

Installation

• Link to PyPi
• Link to Dockerhub

$ pip install lucidtech-synthetic

Make sure to have the following software installed on your system before using the CLI:

• ghostscript

Basic Usage

Docker

We recommend disabling networking and setting /path/to/src_dir to read-only as shown below:

docker run --network none -v /path/to/src_dir:/root/src_dir:ro -v /path/to/dst_dir:/root/dst_dir -it lucidtechai/synthetic pdf /root/src_dir /root/dst_dir
docker run --network none -v /path/to/src_dir:/root/src_dir:ro -v /path/to/dst_dir:/root/dst_dir -it lucidtechai/synthetic image /root/src_dir /root/dst_dir /usr/share/fonts/truetype/dejavu/DejaVuSansMono-Bold.ttf 6-36

CLI

synthetic pdf /path/to/src_dir /path/to/dst_dir
synthetic image /path/to/src_dir /path/to/dst_dir /usr/share/fonts/ubuntu/Ubuntu-B.ttf 6-36

/path/to/src_dir is the input directory and should contain your image/PDFs and JSON ground truths /path/to/dst_dir is the output directory where synthesized image/PDFs and JSON ground truths will be written to

Here is an example of the directory layout for /path/to/src_dir:

/path/to/src_dir
├── a.pdf|jpeg
├── a.json
├── b.pdf|jpeg
└── b.json

The output directory will follow the same layout but with modified images/PDFs and JSON ground truths:

/path/to/dst_dir
├── a.pdf|jpeg
├── a.json
├── b.pdf|jpeg
└── b.json

Using a custom Synthesizer

The following examples shown are for custom PDF synthesizers, but it works similarly for image synthesizers

CLI

synthetic pdf /path/to/src_dir /path/to/dst_dir --synthesizer-class path.to.python.Class

Make sure that parent directory of path.to.python.Class is in your PYTHONPATH

Example using one of the example Synthesizers in examples directory

synthetic pdf /path/to/src_dir /path/to/dst_dir --synthesizer-class examples.exclude-words.synthesizer.ExcludeWordsSynthesizer

Docker

docker run --network none -v /path/to/synthesizer:/root/synthesizer -v /path/to/src_dir:/root/src_dir:ro -v /path/to/dst_dir:/root/dst_dir -it lucidtechai/synthetic pdf /root/src_dir /root/dst_dir --synthesizer-class mypythonfile.ExcludeWordsSynthesizer

Note that the python module must be mounted into the docker container to /root/synthesizer for it to work. In the above example we assume a directory structure of your custom synthesizer to be like below.

/path/to/synthesizer
└── mypythonfile.py

Example using one of the example Synthesizers in examples directory. The examples directory should already exist in the image so that we don't need to mount anything additional.

docker run --network none -v /path/to/src_dir:/root/src_dir:ro -v /path/to/dst_dir:/root/dst_dir -it lucidtechai/synthetic pdf /root/src_dir /root/dst_dir --synthesizer-class examples.exclude-words.synthesizer.ExcludeWordsSynthesizer

Help

All methods support the --help flag which will provide information on the purpose of the method, and what arguments could be added.

$ synthetic --help

Known Issues

Image Synthesizer

• Synthesized text does not follow the rotation of the document in the image if document is rotated
• Bounding boxes needed in ground truth

PDF Synthesizer

• Does not synthesize images inside PDF
• Replaced strings are sometimes not hexadecimal encoded even when expected to be
• Text appearing as single characters with custom spacing in PDF will often yield poor results