Food Packaging OCR Dataset

Description

Food packaging OCR dataset aims to study whether the field-specific data set can improve the text detection and recognition performance of the OCR model on food packaging images. The dataset contains annotation data for text detection and text recognition, covering product name, brand, ingredient list, nutritional information and validity period. The image is collected from the real food packaging and contains different lighting, shooting angles and packaging design conditions, so it can reflect the actual application scenario. This dataset can be used to train, evaluate and compare OCR models.

Steps to reproduce

1. Download and extract the dataset. 2. Verify the det and rec folders are present. 3. Configure the dataset paths in the OCR framework configuration file. 4. Use det/train and det/valid to train and validate a text detection model. 5. Use rec/train and rec/valid to train and validate a text recognition model. 6. Evaluate the trained models using the test splits.

Institutions

• Multimedia University

  Melaka, Malacca

Categories

Licence