jamesnguyen831/fruit-detection-yolov8s

🍎 Fruit Detection with YOLOv8s

A high-accuracy YOLOv8s object detection model trained to detect 6 fruit categories.
This model achieves strong performance and supports real-time inference.

---

📌 Model Overview

• Architecture: YOLOv8s
• Framework: Ultralytics
• Fine-tuned from: yolov8s.pt (COCO pretrained)
• Number of classes: 6
• Training images: 1692
• Validation images: 724
• Image size: 640x640
• Epochs trained: 100
• Hardware: NVIDIA GTX 1650 (4GB VRAM)

---

📊 Performance

Metric	Score
Precision	0.989
Recall	0.974
mAP@0.5	0.992
mAP@0.5:0.95	0.9486

Best checkpoint obtained around epoch 99.

---

🚀 How to Use

Install

pip install ultralytics

Inference on Image

from ultralytics import YOLO

model = YOLO("best.pt")

results = model("test.jpg")
results.show()

Real-Time Webcam

from ultralytics import YOLO

model = YOLO("best.pt")
model.predict(source=0, show=True)

---

🏋️ Training Details

Dataset

• Source: jamesnguyen831/Fruit-Detection-Dataset
• Format: YOLO bounding box format
• 6 fruit categories
• Balanced dataset
• No corrupted images

Training Configuration

• Batch size: 8
• Image size: 640
• Optimizer: Default YOLOv8 optimizer
• Mixed precision: Disabled (AMP off due to GPU compatibility)
• Device: CUDA (GTX 1650)

---

⚠️ Limitations

• Performance may drop under extreme lighting conditions.
• Limited generalization outside trained fruit classes.
• Small dataset size may affect unseen environments.

---

🔬 Technical Specifications

• Parameters: ~11.1M
• GFLOPs: 28.7
• Anchor-free detection head
• CSP-based backbone
• SPPF module

---

🌱 Environmental Impact

Training performed on a single consumer GPU (GTX 1650) for ~5 hours.
Energy consumption is relatively low compared to large-scale models.

---

📚 Citation

If you use this model, please cite YOLOv8:

@software{yolov8_ultralytics,
  author = {Ultralytics},
  title = {YOLOv8},
  year = {2023},
  url = {https://github.com/ultralytics/ultralytics}
}

---

👤 Author

Developed by Quân