Skip to content

A computer vision project (image segmentation project) which aims to remove texts on images using Unet model. Tensorflow 2 is used as a ML library.

Notifications You must be signed in to change notification settings

JiaoPL/DeText

 
 

Repository files navigation

DeText

A computer vision project which aims to remove text on images

It is a ready-to-run code after downloading data from here .

Dependencies

pip3 install -r requirements.txt

Training

Open up a terminal in the main directory and run one of below commands.

If you want to train from scratch:

bash run_from_scratch.sh

If you don't want to train from scratch:

bash run_from_pretrained.sh

Notebook

Notebooks are in notebooks folder.

Data

CelebFaces Attributes (CelebA) Dataset is used in the project.

This is a repository containing more than 200,000 images of various celebrities.

Download the dataset and put the downloaded zip file in main directory. Then, unzip it.

The directories and files should be listed as below:

alt text

Architecture

UNet is used as architecture. I customized it a little. About 2 million Parameters. alt text

Predictions

Predictions for some testing images are below.

alt text

About

A computer vision project (image segmentation project) which aims to remove texts on images using Unet model. Tensorflow 2 is used as a ML library.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%