Instructions to use deni1000/yolo26-text-bubble-detection with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- ultralytics
How to use deni1000/yolo26-text-bubble-detection with ultralytics:
# Couldn't find a valid YOLO version tag. # Replace XX with the correct version. from ultralytics import YOLOvXX model = YOLOvXX.from_pretrained("deni1000/yolo26-text-bubble-detection") source = 'http://images.cocodataset.org/val2017/000000039769.jpg' model.predict(source=source, save=True) - Notebooks
- Google Colab
- Kaggle
| library_name: ultralytics | |
| tags: | |
| - object-detection | |
| - yolo | |
| - manga | |
| - comic | |
| - text-detection | |
| - bubble-detection | |
| metrics: | |
| - map | |
| ## Model Performance | |
| | Metric | Final | Best | | |
| |---|:---:|:---:| | |
| | mAP@50-95 (B) | 0.8434 | 0.8446 | | |
| | mAP@50 (B) | 0.9687 | 0.9865 | | |
| | Precision (B) | 0.9843 | 0.9949 | | |
| | Recall (B) | 0.9511 | 0.9651 | | |
| ### Training Curves | |
| <table> | |
| <tr> | |
| <td align="center"><b>mAP@50-95</b></td> | |
| <td align="center"><b>mAP@50</b></td> | |
| </tr> | |
| <tr> | |
| <td><img src="https://cdn-uploads.huggingface.co/production/uploads/69ae9e7638302cb833f1c09e/HAMMBlZWiWGBgImimpJ_y.jpeg" width="800"></td> | |
| <td><img src="https://cdn-uploads.huggingface.co/production/uploads/69ae9e7638302cb833f1c09e/Odx5ADGSQqghrnqJ7IpL9.jpeg" width="800"></td> | |
| </tr> | |
| <tr> | |
| <td align="center"><b>Precision</b></td> | |
| <td align="center"><b>Recall</b></td> | |
| </tr> | |
| <tr> | |
| <td><img src="https://cdn-uploads.huggingface.co/production/uploads/69ae9e7638302cb833f1c09e/SItT-nCqDEGzSgAgd9P-C.jpeg" width="800"></td> | |
| <td><img src="https://cdn-uploads.huggingface.co/production/uploads/69ae9e7638302cb833f1c09e/disIT1r8L8tURvcl806_y.jpeg" width="800"></td> | |
| </tr> | |
| </table> | |
| ## Training Hyperparameters | |
| ### General | |
| | Parameter | Value | | |
| |---|---| | |
| | Base model | `yolo26s.pt` | | |
| | Task | `detect` | | |
| | Image size | `1280` | | |
| | Epochs | `1200` | | |
| | Batch size | `16` | | |
| | Nominal batch size (nbs) | `64` | | |
| | Device | `1,0` (2× GPU) | | |
| | Workers | `8` | | |
| | Pretrained | `true` | | |
| | AMP | `true` | | |
| | Cache | `true` | | |
| | Seed | `0` | | |
| ### Optimizer | |
| | Parameter | Value | | |
| |---|---| | |
| | Optimizer | `MuSGD` | | |
| | lr0 | `0.01` | | |
| | lrf | `0.005` | | |
| | Cosine LR | `true` | | |
| | Momentum | `0.937` | | |
| | Weight decay | `5e-4` | | |
| | Warmup epochs | `15` | | |
| | Warmup momentum | `0.8` | | |
| | Warmup bias lr | `0.1` | | |
| | Patience (early stop) | `10000` | | |
| ### Loss | |
| | Parameter | Value | | |
| |---|---| | |
| | box | `7.5` | | |
| | cls | `0.5` | | |
| | dfl | `1.5` | | |
| ### Augmentation | |
| | Parameter | Value | | |
| |---|---| | |
| | Mosaic | `0.7` | | |
| | Close mosaic | `30` | | |
| | fliplr | `0.5` | | |
| | flipud | `0.0` | | |
| | translate | `0.1` | | |
| | scale | `0.7` | | |
| | shear | `5` | | |
| | degrees | `0.0` | | |
| | perspective | `3e-4` | | |
| | hsv_h | `0.015` | | |
| | hsv_s | `0.7` | | |
| | hsv_v | `0.4` | | |
| | copy_paste | `0.0` | | |
| | mixup | `0.0` | | |
| | cutmix | `0.0` | | |
| | erasing | `0` | | |
| ### Inference / NMS | |
| | Parameter | Value | | |
| |---|---| | |
| | IoU threshold | `0.7` | | |
| | Max detections | `300` | | |