A curated collection of Jupyter notebooks for experimenting with state-of-the-art OCR, document parsing, table extraction, and chart understanding techniques. This repository enables easy benchmarking and practical usage of the latest open-source and cloud-based solutions for document image processing.
| Notebook | Description |
|---|---|
| bytedance-dolphin-image-parsing.ipynb | Document page parsing with Dolphin by ByteDance |
| Llama-3.1-Nemotron-Nano-VL-8B-V1_parsing_documents.ipynb | Testing the performance of document parsing with Llama-3.1-Nemotron-Nano-VL-8B-V1 |
| docling-documents-parsing-and-tables-extraction.ipynb | Parsing and table extraction with Docling |
| typhoon-ocr-7b-docs-pages-parser.ipynb | Evaluating Typhoon_ocr_7b Document Parsing Capabilities Across Various Use Cases |
| florence-2-large-ocr-documents-pages.ipynb | OCR of document pages using Florence 2 Large |
| florence-2-large-ocr-images-real-life-scenarios.ipynb | Real-life scenario OCR with Florence 2 Large |
| got-ocr2-0-docs-parsing.ipynb | Document pages parsing with GOT-OCR2.0 and Gemini 2.5 Flash |
| marker-docs-parsing.ipynb | Marker-based document parsing experiments |
| mistralocr-docs-parsing.ipynb | Document parsing using MistralOCR |
| monkeyocr-docs-pages-parsing.ipynb | Document parsing with MonkeyOCR |
| nanonets-OCR-s_docs_parsing.ipynb | Advanced document parsing using Nanonets-OCR-s |
| ollama-llama3-2-vision-usage.ipynb | Using Llama3-2 Vision for document parsing |
| paddleocr-3-0-docs-parsing.ipynb | Parsing with PaddleOCR 3.0 PP-StructureV3 |
| pix2text-docs-pages-parsing.ipynb | Document parsing using Pix2Text |
| smoldocling-documents-understanding.ipynb | Document understanding with SmolDocling |
| zerox-pdf-parsing.ipynb | PDF parsing experiments with Zerox |
| qwen2-vl-2b-docs-parsing.ipynb | Documents pages parsing with Qwen2-VL-2B |
| OCRFlux_3B_Docs_Parsing.ipynb | Document parsing with OCRFlux-3B on Lightning AI |
This section includes notebooks focused on table and chart detection, structure recognition, and extraction from documents. It covers various open-source approaches and benchmarks for understanding table and chart layouts and content.
| Notebook | Description |
|---|---|
| unitable-testing-for-table-structure-recognition.ipynb | Testing table detection and structure recognition with UniTable |
| deepdoctection-tables-recognition.ipynb | Evaluating Deepdoctection for table extraction across varied structures |
| gemini-2-5-pro-on-chart-and-table-extraction.ipynb | Chart/table extraction using Gemini 2.5 Pro |
| deplot-plots-to-tables-converter.ipynb | Converting Charts into Tables with DePlot |
This section covers the structured data extraction phase, detailing methods to extract specific data from documents or images. It includes steps like OCR preprocessing, table extraction, named entity recognition (NER), and conversion to structured formats.
| Notebook | Description |
|---|---|
| NuExtract-2-8b-structured-data-extraction | NuExtract-2.0-8B for Structured Data Extraction |
- Benchmark different OCR/document parsing models on real documents.
- Demonstrate table, chart, and text extraction workflows.
- Compare open-source and commercial solutions.
- Provide ready-to-use code snippets for rapid prototyping.
-
Clone the repository:
git clone https://github.com/AdemBoukhris457/Docs_Parsing_Techniques.git
-
Install dependencies as needed for each notebook (see the first cells of each
.ipynbfor requirements). -
Launch Jupyter Notebook or JupyterLab and open any notebook of interest.
-
Run the cells and adapt the code for your documents.
- Some notebooks require model weights or API keys, check comments in each notebook for details.
- Results, insights, and sample outputs are provided inline.
📂 You can find more notebooks, experiments, and datasets related to document parsing and OCR on my Kaggle profile: 👉 https://www.kaggle.com/ademboukhris/code