README.md

metadata

language:
  - en
metrics:
  - accuracy
pipeline_tag: image-text-to-text
base_model:
  - naver-clova-ix/donut-base-finetuned-cord-v2
tags:
  - logistics
  - document-parsing

🏗️ This is a FYP project topic on document parsing of 🚚 logistics 🚚 shipping documents for system integration.

• https://huggingface.co/uartimcs/donut-booking-extract/blob/main/FYP.pdf

Latest update on the version of modules used to continue run the program because there is no recent update for the donut pretrained model.

My use case: Extract common key datafields from shipping documents generated from ten different shipping lines.

Repo & Datasets

• donut.zip (Original Donut Repo + Labelled Booking Dummy Datasets with JSONL files + Config Files)
• sample-image-to-play.zip (Excess dummy datasets used to play and test the model) https://huggingface.co/spaces/uartimcs/donut-booking-gradio

Colab Notebooks

• donut-booking-train.ipynb (Train the model in Colab using T4 TPU / A100 GPU environment)
• donut-booking-run.ipynb (Run the model in Colab using gradio using T4 TPU / A100 GPU environment)

Size of dataset

Follow the CORD-v2 dataset ratio:

• train: 800 (80 pics x 10 classes)
• validation: 100 (10 pics x 10 classes)
• test: 100 (10 pics x 10 classes)