PyTorch-OCR

This project detects and recognizes the text sequences from Chinese shop receipt/bill images.

Note that this is based on the chineseocr and darknet-ocr.

Setup

• Ubuntu 18.04.4

• python 3.7.7

• Install dependencies:

pip install -r requirements.txt

Functions

• Text Detection
• Sequence Recognition
• CPU/GPU Implementation
• Detection Model Training
• Recognition Model Training

Demo

For a simple example, please run:

python main.py

If the 'UnicodeEncodeError' occurs, please run:

PYTHONIOENCODING=utf-8 python main.py

Web Application

• Run the code:

python app.py 8080

• Browse the web page:

  http://127.0.0.1:8080/text

• Click the '上传本地照片' button to upload a receipt/bill image. Click '识别' to recognize it.

Results

References

• We convert the chineseocr that is implemented by Darknet framework to PyTorch.
• The web application setup refers to darknet-ocr.