|
|
--- |
|
|
license: mit |
|
|
base_model: |
|
|
- Ultralytics/YOLOv8l |
|
|
--- |
|
|
## Model Training |
|
|
|
|
|
### Training Details |
|
|
|
|
|
The YOLOv8l model was fine-tuned on a **cloud A100 GPU** (NVIDIA A100-SXM4-40GB) using approximately **24,000 images** from the Augmented Startups Playing Cards dataset. |
|
|
|
|
|
#### Training Configuration: |
|
|
- **Model**: YOLOv8l (YOLO v8 Large) |
|
|
- **Dataset**: Augmented Startups Playing Cards (Roboflow Universe) |
|
|
- Dataset URL: https://universe.roboflow.com/augmented-startups/playing-cards-ow27d/dataset/4 |
|
|
- **Training Images**: ~24,000 images |
|
|
- **Classes**: 52 (one for each playing card) |
|
|
- **Epochs**: 50 |
|
|
- **Image Size**: 640x640 |
|
|
- **Batch Size**: 16 |
|
|
- **Hardware**: NVIDIA A100-SXM4-40GB GPU |
|
|
- **Framework**: Ultralytics YOLOv8 |
|
|
|
|
|
#### Training Process: |
|
|
|
|
|
The training was performed using the Ultralytics YOLOv8 framework. The process involved: |
|
|
|
|
|
1. **Dataset Preparation**: Downloaded the Augmented Startups Playing Cards dataset from Roboflow in YOLOv8 format |
|
|
2. **Model Initialization**: Started with pre-trained YOLOv8l weights (`yolov8l.pt`) |
|
|
3. **Fine-tuning**: Trained for 50 epochs on the playing cards dataset |
|
|
4. **Model Export**: Saved the fine-tuned model as `playing-cards.pt` |
