# DER (Dynamic Enhancement for Object Detection)

This repository contains trained model weights for the DER (Dynamic Enhancement for object detection) project.

## Repository Structure

```
Model/
├── baseline/              # Baseline model weights
│   ├── VisDrone2019/
│   ├── UAVDT/
│   ├── TinyPerson/
│   └── DOTAv1/
└── DER_improved/          # DER-enhanced model weights
    ├── VisDrone2019/
    ├── UAVDT/
    ├── TinyPerson/
    └── DOTAv1/
```

## Supported Models

- **RTMDet-R2**: Real-time object detector with rotated bounding boxes
  - Scales: tiny, small
  - Format: `.pth` (PyTorch)
- **PP-PicoDet**: Lightweight object detection model from PaddlePaddle
  - Scales: m (medium), l (large)
  - Format: `.pdparams` (PaddlePaddle)
- **YOLOv11**: (Coming soon)
- **RT-DETR**: (Coming soon)

## Datasets

- **VisDrone2019**: Drone-based object detection dataset
- **UAVDT**: UAV-based detection and tracking dataset
- **TinyPerson**: Small object detection dataset
- **DOTAv1**: Dataset for Object deTection in Aerial images

## File Naming Convention

Files are named following the pattern: `ModelName-Scale-Dataset.ext`

Examples:
- `PP-PicoDet-l-VisDrone2019.pdparams`
- `RTMDet-R2-tiny-UAVDT.pth`

## Available Weights

### PP-PicoDet
- **baseline**: 8 weight files (2 scales × 4 datasets)
- **DER_improved**: 8 weight files (2 scales × 4 datasets)

### RTMDet-R2
- **baseline**: 8 weight files (2 scales × 4 datasets)
- **DER_improved**: 8 weight files (2 scales × 4 datasets)

## Usage

```python
from huggingface_hub import hf_hub_download

# Download PP-PicoDet baseline weights
weight_path = hf_hub_download(
    repo_id="Nahuyiur/DER",
    filename="PP-PicoDet/baseline/VisDrone2019/PP-PicoDet-l-VisDrone2019.pdparams"
)

# Download RTMDet-R2 DER-improved weights
weight_path = hf_hub_download(
    repo_id="Nahuyiur/DER",
    filename="RTMDet-R2/DER_improved/TinyPerson/RTMDet-R2-small-TinyPerson.pth"
)
```

## License

Please refer to the original model repositories for license information.