Shape Mining Pipeline

Description

The shape mining pipeline is designed to extract veselprofiles with corresponding metadata from book scans.

1. Extract figureID and pageID from book scan
2. Extract veselprofiles from book scan
3. Extract only characteristic shape of the veselprofiles
4. kNN approach to find most similar vesel shapes

Getting Started

There are two Docker Containers. One for a GPU machine and one for a CPU machine. Install and run docker container. Container hosts a jupyter server. Ther URL for accessing the server will be shown in the terminal.

chmod +x start_docker.sh
./start_docker.sh

or use docker-compose for your preferred config (CPU or GPU). Example for CPU

docker-compose -f Docker_CPU/docker-compose.yml up

Access Docker shell. The container has to run for that.

docker ps (to check COTAINER_ID)
docker exec -it CONTAINER_ID bash

Development with Visual Studio Code

1. Install Visual Studio Code with the following extensions:

• Remote Development
• Docker

2. Open devcontainer file and choose in entry "dockerComposeFile" the GPU or CPU container.
3. Create following directories if you use a Linux OS:

• vscode_remote/extensions
• vscode_remote/bashhistory
• vscode_remote/insiders

4. In VSCodee press Shift+P and run "Remote-Containers:Rebuild and Reopen in Container" command.

Cotainer filesystem

Source code is located at /home/Code
Tensorflow objection detection API at /models/research/object_detection

Models

Models for mining shapes can be downloaded at Mining Pages

Run whole pipeline

Mount correct volumes to docker-compose file
Run file mining_pages.py