README.md

---
tags:
- yolo
- yolov8
- segmentation
- overlay-detection
- computer-vision
- instance-segmentation
library_name: ultralytics
license: agpl-3.0
---

# YOLO Overlay Detection Model

This model was trained to detect and segment overlay elements in images/videos using YOLOv8 segmentation.

This repository contains two primary model files:
- `best.pt`: The model checkpoint with the best validation metrics seen so far.
- `last.pt`: The final checkpoint from the most recent training run, used for resuming.

## Model Details

- **Model Type**: YOLOv8 Instance Segmentation
- **Architecture**: yolov8m-seg
- **Framework**: Ultralytics YOLO
- **Training Date**: 2025-11-04
- **Task**: Instance Segmentation
- **Classes**: Overlay elements

## Performance Metrics (from last 'best.pt')

| Metric | Value |
|--------|-------|
| Box mAP@0.5 | 0.9093 |
| Box mAP@0.5:0.95 | 0.7576 |
| Mask mAP@0.5 | 0.6030 |
| Mask mAP@0.5:0.95 | 0.2714 |


## Usage

### Installation

```bash
pip install ultralytics
```

### Inference (Using the best model)

```python
from ultralytics import YOLO
from huggingface_hub import hf_hub_download

# Download best model
model_path = hf_hub_download(
    repo_id="farazv2/overlay-model-yolo",
    filename="best.pt"
)

# Load model
model = YOLO(model_path)

# Run inference
results = model('image.jpg')
```

### Resuming Training

```python
from ultralytics import YOLO
from huggingface_hub import hf_hub_download

# Download last model
model_path = hf_hub_download(
    repo_id="farazv2/overlay-model-yolo",
    filename="last.pt"
)

# Load model and resume
model = YOLO(model_path)
model.train(data='path/to/data.yaml', resume=True)
```

## Training Configuration

| Parameter | Value |
|-----------|-------|
| Epochs | 10 (per run) |
| Image Size | 640 |
| Optimizer | AdamW |
| Initial Learning Rate | 0.001 |
| Batch Size | 24 |
| Mixed Precision | True |
| Patience | 20 |


## License

This model is released under the AGPL-3.0 license, following Ultralytics YOLOv8 licensing.