OCR Indonsia

This repositories is for prototype OCR's API deployment.

Testing Environment

ROG Strix Hero II GL504GV
NVIDIA GeForce RTX 2060
=========================
Ubuntu 18.04
=========================
python 3.8.3
cuda 10.2
pytorch 1.5.1
opencv 4.3.0

Preparation

Download and Install the mandatory packages, using script below :

conda install python
conda install pytorch torchvision cudatoolkit=10.2 -c pytorch
pip install opencv-contrib-python
pip install pandas

Build Detectron2 localy, you can refer to this link but here we will give you a quick installation :

Windows

python -m pip install 'git+https://github.com/facebookresearch/detectron2.git'

Linux

python -m pip install detectron2 -f \
  https://dl.fbaipublicfiles.com/detectron2/wheels/cu102/torch1.5/index.html

Clone this repository git clone https://gitlab.com/aempat/proto_ocr.git
Download our pretrained model here and here, and put it in model_weights folder.

if you havn't downloaded the model yet. We make the model can be downloaded automatically, when you call the module

├── model_weights
│   ├── faster_rcnn_R_101_FPN_3x_NIK_NOSIM.pth
│   │   ...
│   └── model_final_ktp_sim-fix_rgb_rotated_X101_custom_mapper.pth
└── Examples.ipynb

Demo

We provide streamlit apps for this OCR application, follow this 2 steps in terminal inside this repo:

pip install streamlit
streamlit run streamlit_ocr.py

go to http://{your ip or localhost}:8501, select document type, upload photo then get the result

Detail Example

you can run our notebook for detailed example. Please be aware with list_cuda parameters on cfg, the default is set on 4 GPUs. if you only have one GPU, set list_cuda to 0 in every cfg.

Contact

[Ziyad S Fawwazi]
[Hervind]
[Andreas Mulya R]

vindruid/ocr_indonesia