YOLOv8m Table Extraction
keremberke/yolov8m-table-extraction
published Jan 2023 · updated May 2024
YOLOv8m Table Extraction is a detection model that identifies bordered and borderless tables in images.
specs
| Task | Object Detection (Table Extraction) |
| Architecture | YOLOv8m (YOLOv8 medium variant) |
| License | AGPL-3.0 |
| Performance | [email protected] (box) = 0.952 on table-extraction validation set |
about this model
keremberke/yolov8m-table-extraction is a detection model that identifies and classifies tables in images as either "bordered" or "borderless". It is built on the YOLOv8m architecture and is hosted as a managed, OpenAI-compatible API by gigarouter.
Capabilities and Performance
The model is trained on the keremberke/table-extraction dataset, which contains 342 images (238 training, 70 validation, 34 test). The dataset was originally sourced from Roboflow and covers two table styles: bordered (category 0) and borderless (category 1). On the validation set, the model achieves a self-reported mean average precision at IoU 0.5 ([email protected]) of 0.952 for bounding boxes.
Key Strengths
- High accuracy: [email protected] of 0.952 on the validation set, indicating reliable detection and classification of table types.
- Lightweight specialized model: As a YOLOv8m variant, it balances inference speed and precision for table extraction tasks.
- Table-style awareness: Differentiates between bordered and borderless tables, enabling context‑appropriate downstream processing.
Additional Details
The model is released under the AGPL‑3.0 license and has accumulated over 176,000 monthly downloads on Hugging Face, reflecting its practical utility in document analysis and data extraction workflows.
Because gigarouter hosts the model as a managed API, you can integrate it into your application with a single API call — no local installation or dependency management required.
best for
- ·Extracting tables from scanned documents or images
- ·Automating data entry from forms containing tables
- ·Detecting both bordered and borderless table regions in digital documents
FAQ
It detects tables in images, classifying them as either 'bordered' or 'borderless'.
Send a POST request to the gigarouter OpenAI-compatible endpoint with your API key, providing an image URL or base64 encoded image data.
The model is licensed under AGPL-3.0.
It achieves a mean average precision ([email protected]) of 0.952 on the table-extraction validation set.
The model accepts standard image formats (JPG, PNG, etc.) via URL or base64 encoding when using the gigarouter API.
We're benchmarking and onboarding YOLOv8m Table Extraction as a hosted, OpenAI-compatible API. Sign in for free credit and be ready when it lands, or tell us you want it and we'll prioritize it.