|
|
--- |
|
|
license: mit |
|
|
tags: |
|
|
- yolo |
|
|
- object-detection |
|
|
- knowledge-distillation |
|
|
- quantization |
|
|
- military |
|
|
- tank-detection |
|
|
datasets: |
|
|
- roboflow/military-object-detection |
|
|
metrics: |
|
|
- mAP |
|
|
--- |
|
|
|
|
|
# YOLO Tank Detection Models |
|
|
|
|
|
Military object detection models trained with Knowledge Distillation and Post-Training Quantization. |
|
|
|
|
|
## Models |
|
|
|
|
|
| Model | mAP50 | mAP50-95 | Size | |
|
|
|-------|-------|----------|------| |
|
|
| Teacher (YOLOv8m) | 86.40% | 62.78% | 49.6 MB | |
|
|
| Student Fine-tuned (YOLOv8n) | 84.61% | 60.64% | 5.96 MB | |
|
|
| Student KD (YOLOv8n) | 79.28% | 53.38% | ~6 MB | |
|
|
| Student Quantized (INT8) | 84.31% | 60.50% | 3.20 MB | |
|
|
|
|
|
## Usage |
|
|
|
|
|
```python |
|
|
from huggingface_hub import hf_hub_download |
|
|
|
|
|
# Download quantized model for web deployment |
|
|
model_path = hf_hub_download( |
|
|
repo_id="Hunjun/yolo-tank-detection", |
|
|
filename="optimized/student_quantized.onnx" |
|
|
) |
|
|
|
|
|
# Or download teacher model |
|
|
teacher_path = hf_hub_download( |
|
|
repo_id="Hunjun/yolo-tank-detection", |
|
|
filename="teacher/yolov8m_tank_best.pt" |
|
|
) |
|
|
``` |
|
|
|
|
|
## Classes |
|
|
|
|
|
- Airplane |
|
|
- Helicopter |
|
|
- Person |
|
|
- Tank |
|
|
- Vehicle |
|
|
|
|
|
## Training |
|
|
|
|
|
- Teacher: YOLOv8m fine-tuned on Military Dataset (20 epochs) |
|
|
- Student: YOLOv8n fine-tuned / Knowledge Distillation (20 epochs) |
|
|
- Quantization: INT8 dynamic quantization via ONNX Runtime |
|
|
|
|
|
## References |
|
|
|
|
|
- [Ultralytics YOLOv8](https://github.com/ultralytics/ultralytics) |
|
|
- [Knowledge Distillation (Hinton et al., 2015)](https://arxiv.org/abs/1503.02531) |
|
|
|
