peternara/ocr-2

Scene-OCR: CRAFT: text detection + TPS-ResNet-BiLSTM-Attn: text recognition

Scene-OCR

CRAFT: text detection + TPS-ResNet-BiLSTM-Attn: text recognition

This repo combines two methods which are most popular and having best performace to meet my demand in Scene-OCR. I have modified two forked repos and rewrite a Flask service for local users to use detection, recognition and OCR conveniently.

How to use

1. Install all dependency. If there are still some missing packages, you can check original forked repos.

pip install -r requirements.txt

2. It is better to run app under GPU.

CUDA_VISIBLE_DEVICES=0 python app.py

3. Try it by yourself. You can send POST requests to http://localhost:8301/ocr with your image absolute path. See more details in test_app.py.

Text Detection

This part is forked by clovaai/CRAFT-pyotrch.

Loaded model is models/craft_mlt_25k.pth. You can download it from here.

Text Recognition

This part is forked by clovaai/deep-text-recognition-benchmark

Loaded model is models/TPS-ResNet-BiLSTM-Attn.pth. You can download it from here

Notice

The link for both models are supported by original repos. If there are any issues, you could post issues in there.