/document-layout-analysis-app

Simple docker deployment of document layout analysis using detectron2

Primary LanguageJavaScriptMIT LicenseMIT

Document Layout Analysis App

Simple docker deployment of document layout analysis using detectron2. Trained on Publaynet dataset.

Backend is built in Flask

Frontend is built in React

Steps to build

  • Run docker-compose up --build from the main folder
  • Mask RCNN R101 will be auto downloaded
  • Type localhost:3000 in browser and upload and image

(Optional) To use model other than Mask RCNN R101

  • Download pre-trained model refered in https://github.com/hpanwar08/detectron2 (default Resnet101 model can be downloaded from MaskRCNN Resnet101 FPN 3X )
  • Copy the downloaded model in backend folder
  • Download config yaml refered in https://github.com/hpanwar08/detectron2 (default Resnet101 config yaml is already present in backend folder)
  • Copy the downloaded config yaml in backend folder
  • Update the model_config.yaml file with model file name and config yaml file name (not required if using default values)
  • Run docker-compose up --build from the main folder
  • Type localhost:3000 in browser and upload and image

Note: It takes time to build the image