| --- |
| license: agpl-3.0 |
| library_name: ultralytics |
| pipeline_tag: object-detection |
| tags: |
| - yolov8 |
| - computer-vision |
| - object-detection |
| - technical-catalogs |
| - industrial-automation |
| - hotspot-detection |
| --- |
| |
| # Machine Catalog Hotspot Detector |
|
|
| A custom-trained YOLOv8s object detection model developed to detect hotspot regions in industrial machinery technical catalogs. |
|
|
| ## Overview |
|
|
| Industrial machinery catalogs often contain complex exploded-view diagrams, numbered components, and regions of interest used for spare parts identification. |
|
|
| This project explores the use of deep learning to automatically detect hotspot regions within technical catalog pages, supporting automated catalog digitization and downstream document processing workflows. |
|
|
| ## Model Information |
|
|
| | Property | Description | |
| |---|---| |
| | Architecture | YOLOv8s | |
| | Framework | Ultralytics / PyTorch | |
| | Task | Object Detection | |
| | Detection Class | Hotspot | |
| | Model Format | PyTorch (.pt) | |
| | Base Model | yolov8s.pt | |
|
|
| ## Intended Applications |
|
|
| - Automated hotspot detection in technical catalog pages |
| - Industrial spare parts catalog digitization |
| - Technical document analysis |
| - Computer vision research and experimentation |
| - Preprocessing for industrial parts identification systems |
|
|
| ## Quick Start |
|
|
| Install the required dependency: |
|
|
| ```bash |
| pip install ultralytics |
| ``` |
|
|
| Load the model directly from Hugging Face: |
|
|
| ```python |
| from ultralytics import YOLO |
| from huggingface_hub import hf_hub_download |
| |
| model_path = hf_hub_download( |
| repo_id="suleymankolenoglu/MachineCatalogHotspotDetector", |
| filename="best.pt" |
| ) |
| |
| model = YOLO(model_path) |
| |
| results = model.predict( |
| source="catalog_page.jpg", |
| conf=0.25, |
| save=True |
| ) |
| ``` |
|
|
| ## Training |
|
|
| The model was fine-tuned from the pretrained YOLOv8s architecture for a custom single-class hotspot detection task. |
|
|
| ## Evaluation |
|
|
| The model was tested during development on technical catalog images. |
|
|
| Formal evaluation metrics have not yet been published. Quantitative benchmarks may be added in future updates. |
|
|
| ## Limitations |
|
|
| - Performance may vary across different catalog layouts and image resolutions. |
| - Small or visually ambiguous hotspot regions may be difficult to detect. |
| - Generalization to previously unseen catalog styles has not been systematically evaluated. |
| - The model is not intended for safety-critical applications. |
|
|
| ## Future Improvements |
|
|
| - Publish quantitative evaluation results |
| - Add example predictions |
| - Develop an interactive Hugging Face Spaces demonstration |
| - Evaluate generalization across different technical catalog formats |
|
|
| ## License |
|
|
| AGPL-3.0. Use and redistribution are subject to applicable Ultralytics licensing terms. |
|
|
| ## Acknowledgments |
|
|
| Built using the Ultralytics YOLOv8 framework. |
