About the project

This project was developed for the Vision and Perception course for the academic year 2022-2023 at Sapienza University of Rome. The goal is to remove text from images. To achieve this goal, two different neural networks have been implemented.

• UNet: to obtain masks covering the text in the images
• TextAwareMultiGan: to impainting
  The weights of the networks we obtained are available at the link: https://drive.google.com/drive/folders/10KcpzptTwK6jJJdPxIUwdt1X2hTUj2_Z?usp=drive_link

Getting Started

Installation

1. Clone the repo
   git clone https://github.com/paolacarboni/project-vision-perception.git
2. Install Python packages
   2.1 Linux/MacOs
   make install
   2.2 Windows
   install all the packages in the requirements file

Usage

Training

• TextAwareMultiGan
  • Run make dataset by selecting a folder structured as follows:
    train
    |-- masks
    |-- textures
    test
    |-- masks
    |-- textures
  • Run make train and select gan
• UNet Run make train and select text

Testing

You can launch an interface that allows text detection, inpainting, or text removal by using make exec. To do this, include the networks weights in the "resources/neuralNetworks" folder.