Skip to content

iuliaturc/detextify

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

11 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Detextify

What is this?

TL;DR: A Python library to remove unwanted pseudo-text from images generated by your favorite generative AI models (Stable Diffusion, Midjourney, DALL·E).

Before After
before after

So, why should I care?

We all know generative AI is the coolest thing since sliced bread 🍞.

But try using any off-the-shelf generative vision model and you'll quickly see that these systems can get... creative with interpreting your prompts.

Specifically, you'll observe all kinds of weird artifacts on your images from extra fingers on hands, to arms coming out of chests, to alien text written in random places.

For generative systems to actually be usable in downstream applications, we need to better control these outputs and mitigate unwanted effects.

We believe the next frontier for generative AI is about robustness and trust. In other words, how can we architect these systems to be controllable, relevant, and predictably consistent with our needs?

Detextify is the first phase in our vision of robustifying generative AI.

If we get this right, we will unlock slews of new applications for generative systems that will change the landscape of human-AI collaboration. 🌎

Cute, but what are you actually doing?

Detextify runs text detection on your image, masks the text boxes, and in-paints the masked regions until your image is text-free. Detextify can be run entirely on your local machine (using Tesseract for text detection and Stable Diffusion for in-painting), or can call existing APIs (Azure for text detection and OpenAI or Replicate for in-painting).

Installation

pip install detextify

Additionally:

  • To run text detection locally (as opposed to using the Azure API), you need to install Tesseract.
  • To run in-painting locally (as opposed to using the OpenAI or Replicate APIs), you need a GPU with CUDA and cuDNN installed.

Usage

See this Colab notebook for how to use the library, or follow the instructions below.

You can remove unwanted text from your image in just a few lines 💪:

from detextify.text_detector import TesseractTextDetector
from detextify.inpainter import LocalSDInpainter
from detextify.detextifier import Detextifier

text_detector = TesseractTextDetector("/path/to/tesseract/installation")
detextifier = Detextifier(text_detector, LocalSDInpainter())
detextifier.detextify("/my/input/image/path.png", "/my/output/image/path.png")

and 💣💥, just like that, your image is cleared of any bizarre text artifacts.

Or if you want to clean up a directory of PNG images, just wrap it in a for-loop:

import glob
from detextify.text_detector import TesseractTextDetector
from detextify.inpainter import LocalSDInpainter
from detextify.detextifier import Detextifier

text_detector = TesseractTextDetector("/path/to/tesseract/installation")
detextifier = Detextifier(text_detector, LocalSDInpainter())
for img_file in glob.glob("/path/to/dir/*.png"):
    detextifier.detextify(img_file, img_file.replace(".png", "_detextified.png"))

We provide multiple implementations for text detection and in-painting (both local and API-based), and you are also free to add your own.

Text Detectors

  1. TesseractTextDetector (based on Tesseract) runs locally. Follow this guide to install the tesseract library locally. On Ubuntu:
sudo apt install tesseract-ocr
sudo apt install libtesseract-dev

To find the path where it was installed (and pass it to the TesseractTextDetector constructor):

whereis tesseract
  1. AzureTextDetector calls a computer vision API from Microsoft Azure. You will first need to create a Computer Vision resource via the Azure portal. Once created, take note of the endpoint and the key.
AZURE_CV_ENDPOINT = "https://your-endpoint.cognitiveservices.azure.com"
AZURE_CV_KEY = "your-azure-key"
text_detector = AzureTextDetector(AZURE_CV_ENDPOINT, AZURE_CV_KEY)

Our evaluation shows that the two text detectors produce comparable results.

In-painters

  1. LocalSDInpainter (implemented via Huggingface's diffusers library) runs locally and requires a GPU. Defaults to Stable Diffusion v2 for in-painting.
  2. ReplicateSDInpainter calls the Replicate API. Defaults to Stable Diffusion v2 for in-painting (and requires an API key).
  3. DalleInpainter calls the DALL·E 2 API from OpenAI (and requires an API key).
# You only need to instantiate one of the following:
local_inpainter = LocalSDInpainter()
replicate_inpainter = ReplicateSDInpainter("your-replicate-key")
dalle_inpainter = DalleInpainter("your-openai-key")

Contributing

To contribute, clone the repository, make your changes, commit and push to your clone, and submit a pull request.

To build the library, you need to install poetry:

curl -sSL https://install.python-poetry.org | python3 -
# Add poetry to your PATH. Note the specific path will differ depending on your system.
export PATH="/home/ubuntu/.local/bin:$PATH"
# Check the installation was successful:
poetry --version

Install dependencies for detextify:

poetry install

To execute a script, run:

poetry run python your_script.py

Please run the unit tests to make sure that your changes are not breaking the codebase:

poetry run pytest

Authors

This project was authored by Mihail Eric and Julia Turc. If you are building in the generative AI space, we want to hear from you!