Handwriting Recognition

Dead Sea Scrolls Task 1 & 2

• The code was built in Python version 3.8.13

Preparatory steps

• Please follow these steps in the specified order
• Make sure you have a working installation of the latest R (https://cran.r-project.org/bin/windows/base/) - an R version of at least 4.1.3 is required for installing and running the packages used
• Ensure you have python(3) installed on your system.
• Change directory to the root directory of this project. On linux, this can be done through:

cd path/to/folder_with_task1&2/located/somewhere

• Create the virtual environment as follows:

python3 -m venv venv # creates a virtualenv (venv) where dependencies will be installed
source venv/bin/activate # activates the virtualenv
pip3 install -r requirements.txt # installs the dependencies

• Executing the code above will install the Tensorflow package on your system. If you wish to run Tensorflow from GPU: the model was compiled using cuDNN 8.1.0 and Cuda 11.2, so please make sure you have these installed as well (conda example for installation: conda install -c conda-forge cudatoolkit=11.2 cudnn=8.1.0)
• Please make sure that you have the test images located inside of a folder

Running the model

• Please make sure that you're still located (terminal) in the root directory of Task 1 & 2.
• Run main.py in the format:

python3 main.py path/to/folder_with_jpg/located/somewhere

• The folder should contain jpg images of binarized Dead Sea Scrolls
• Final txt files can be found in the results folder

Extra scripts

The folder additional_code/ includes some of the code that was written in the process of investigating the problem, trying out different pipelines and models, or simply for convenience. These were not used in the final pipeline, but may be nice to take a look at.

Citation

Line segmentation adapted from:

@inproceedings{Surinta2014Astar,
	author = {Surinta, Olarik and Holtkamp, Michiel and Karaaba, Mahir and Oosten, Jean-Paul and Schomaker, Lambert and Wiering, Marco},
	year = {2014},
	month = {05},
	pages = {},
	title = {A* Path Planning for Line Segmentation of Handwritten Documents},
	volume = {2014},
	journal = {Proceedings of International Conference on Frontiers in Handwriting Recognition, ICFHR},
	doi = {10.1109/ICFHR.2014.37}
}

Model adapted from

https://keras.io/examples/vision/captcha_ocr/

IAM Task 3

Installation

1. Since you are reading this README, we will assume that you have managed to download the content locally. If not, please download/clone it.
2. Change directory to the root directory of this project. On linux, this can be done through:

cd path/to/folder/located/somewhere

3. List the current files in the directory, and make sure you see files such as create_lmdb_dataset.py, demo.py, run.py, ..., etc. If not, please retry step 2. On linux, this can be done through:

ls -A

4. Ensure you have python(3) installed on your system.
5. Create the virtual environment as follows:

python3 -m venv venv # creates a virtualenv (venv) where dependencies will be installed
source venv/bin/activate # activates the virtualenv
pip3 install -r requirements.txt # installs the dependencies

NOTE: The current requirements file will install the CPU version of PyTorch, since not all systems are CUDA-enabled, and the demo will run just fine on the CPU. If you prefer to install a CUDA-enabled version of PyTorch, remove these entries from the requirements file. After which you can follow the installation instructions here.

Prerequisites

1. You have the test images located inside of a folder, e.g.:

- <root-project-directory>/
    - <test-images-directory>/
        - <image-1>.png
        - <image-2>.png
        - ...
        - <image-N>.png

2. You have followed the installation instructions, and are still located (terminal) in the root directory of the project.
3. You have noted the name of the directory which holds your test images.

Run instructions

1. Ensure you meet the prerequisites described above.
2. To run the demo, execute (replacing the test image directory to your specification):

python3 run.py --folder <test-images-directory>

3. The predictions will have been written to results/*.txt.

Extra scripts

The folder additional_code/ includes some of the code that was written in the process of investigating the problem, trying out different pipelines and models, or simply for convenience. These were not used in the final pipeline, but may be nice to take a look at. The additional code includes processes such as binarization, smart padding and resizing, encoding and decoding labels, splitting up the original training data (also for k-fold cross-validation), defining a CRNN, creating a PyTorch dataset out of the images and ground-truths.

Citation

Codebase partially adapted from:

@inproceedings{baek2019STRcomparisons,
  title={What Is Wrong With Scene Text Recognition Model Comparisons? Dataset and Model Analysis},
  author={Baek, Jeonghun and Kim, Geewook and Lee, Junyeop and Park, Sungrae and Han, Dongyoon and Yun, Sangdoo and Oh, Seong Joon and Lee, Hwalsuk},
  booktitle = {International Conference on Computer Vision (ICCV)},
  year={2019},
  note={to appear},
  pubstate={published},
  tppubtype={inproceedings}
}

License

Copyright (c) 2019-present NAVER Corp.

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.