This repository implements YOLOv11n for dental image analysis, featuring a lightweight architecture optimized for efficient inference while maintaining high detection accuracy. The model utilizes the Ultralytics framework and incorporates advanced features like C3K2 blocks and SPPF (Spatial Pyramid Pooling - Fast).
The network consists of 319 layers with 2,590,035 parameters (2,590,019 gradients) and operates at 6.4 GFLOPs. Key architectural components include:
- Input Resolution: 640x640
- Backbone: Progressive feature extraction using Conv and C3K2 blocks
- Neck: Feature pyramid with upsampling and concatenation
- Head: Multi-scale detection with 64x64, 128x128, and 256x256 feature maps
- Initial Convolutions:
3→16→32channels - C3K2 blocks with varying channel depths:
64→128→256 - SPPF module with 256 channels and 5x5 maximum pooling
- C2PSA attention module
- Detection head with three scales
The model was trained for 5 epochs with the following metrics:
-
Initial (Epoch 1/5):
- Box Loss:
2.55 - Classification Loss:
1.526 - DFL Loss:
1.526 - mAP50:
0.238 - mAP50-95:
0.154
- Box Loss:
-
Final (Epoch 5/5):
- Box Loss:
1.533 - Classification Loss:
1.513 - DFL Loss:
1.404 - mAP50:
0.665 - mAP50-95:
0.401
- Box Loss:
pip install ultralyticsyolo detect train model=yolov11n.pt data=data.yaml epochs=5yolo detect predict model=path/to/best.pt source=path/to/imagesThe model demonstrates progressive improvement across training epochs:
- mAP50 increased from
0.238to0.665 - mAP50-95 improved from
0.154to0.401 - Processing speed: ~4-6 seconds per iteration at 640x640 resolution
- Python 3.8+
- PyTorch
- Ultralytics
- CUDA-capable GPU (recommended)
This project uses the Ultralytics YOLO framework and follows its licensing terms.
- Ultralytics for the YOLOv11n architecture
- Original YOLO authors for the detection methodology