Semantic Segmentation of Historical Documents using Deep Learning Architectures
- Prediction of Semantic Segmentation masks for images of the Eparchos Dataset.
- Deep Learning Architectures used: dhSegment, U-Net, VGG16
- This consists a demo of my Graduate Thesis.
1.It is recommended to install tensorflow (or tensorflow-gpu) independently using Anaconda distribution, in order to make sure all dependencies are properly installed.
2.Clone the repository using git clone https://github.com/anastzel/hist-doc-segm.git
3.Install Anaconda or Miniconda
4.Create a virtual environment and activate it
conda create -n hist_doc_segm python=3.6
source activate hist_doc_segm
5.Install dhSegment dependencies with pip install -r requirements.txt
6.Install TensorFlow 1.13 with conda conda install tensorflow-gpu=1.13.1
You can download the models needed from my here (models have a large size, so GitHub didn't allow me to upload them).
After you have downloaded the models, place them inside the repository folder.
To run the Demo open Anaconda prompt and navigate to the repository folder.
Then run streamlit run appication.py.
A browser's tab should open containing the application's demo.
