YOLO11x – Fine-tuned Watermark Detector
This is a fine-tuned YOLO11x model for watermark & logo detection, built upon the original fancyfeast/joycaption-watermark-detection model (by fancyfeast). I further trained it on a custom dataset for improved performance.
Model Details
| Feature | Value |
|---|---|
| Base checkpoint | fancyfeast/joycaption-watermark-detection |
| Input size | 1280 × 1280 |
| Framework | PyTorch / Ultralytics YOLO11x |
Installation
pip install ultralytics opencv-python
Usage
Python
from ultralytics import YOLO
import cv2
# Load model
model = YOLO("corzent/yolo11x_watermark_detection")
# Run inference
results = model("test.jpg", conf=0.35)
# Visualize
annotated = results[0].plot()
cv2.imwrite("result.jpg", annotated)
CLI
yolo predict model=your-username/wm_yolo11_ft source="test.jpg" conf=0.3
Results
| Metric | Value |
|---|---|
| Precision | 0.896 |
| Recall | 0.883 |
| mAP@50 | 0.900 |
| mAP@50-95 | 0.686 |
Model Info
- Base model: YOLO11x
- Trained on: custom watermark/logo dataset
License
This model is provided for research and educational purposes only.
Do not use for removing watermarks from copyrighted content.
🤝 Created by Corzent