File size: 11,059 Bytes
c46f468 2910c41 c46f468 2910c41 c46f468 2910c41 c46f468 2910c41 c46f468 2910c41 c46f468 2910c41 c46f468 2910c41 c46f468 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 | ---
license: cc-by-nc-4.0
task_categories:
- object-detection
---
# HazyDet: Open-Source Benchmark for Drone-View Object Detection With Depth-Cues in Hazy Scenes [(paper)](https://arxiv.org/abs/2409.19833)
**HazyDet** is the first benchmark for object detection in hazy drone imagery. It couples physics-driven synthetic data with real foggy drone photos, providing a controlled yet realistic test-bed for designing haze-robust detectors.
---
## Abstract
Object detection from aerial platforms under adverse atmospheric conditions, particularly haze, is paramount for robust drone autonomy.
Yet, this domain remains largely underexplored, primarily hindered by the absence of specialized benchmarks.
To bridge this gap, we present HazyDet, the first, large-scale benchmark specifically designed for drone-view object detection in hazy conditions.
Comprising 383,000 real-world instances derived from both naturally hazy captures and synthetically hazed scenes augmented from clear images,
HazyDet provides a challenging and realistic testbed for advancing detection algorithms. To address the severe visual degradation induced by haze,
we propose the Depth-Conditioned Detector (DeCoDet), a novel architecture that integrates a Depth-Conditioned Kernel to dynamically modulate feature representations
based on depth cues. The practical efficacy and robustness of DeCoDet are further enhanced by its training with a Progressive Domain Fine-Tuning (PDFT) strategy
to navigate synthetic-to-real domain shifts, and a Scale-Invariant Refurbishment Loss (SIRLoss) to ensure resilient learning from potentially noisy depth annotations.
Comprehensive empirical validation on HazyDet substantiates the superiority of our unified DeCoDet framework,
which achieves state-of-the-art performance, surpassing the closest competitor by a notable +1.5\% mAP on challenging real-world
hazy test scenarios. Our dataset and toolkit are available at [github](https://github.com/GrokCV/HazyDet).
## HazyDet
---
### 📦 Dataset at a Glance
_Target size buckets: Small < 0.1 % of image area , Medium 0.1–1 % , Large > 1 %_
| Split | #Images | #Instances | Class | Small | Medium | Large |
|-------|:-------:|:----------:|-------|------:|-------:|------:|
| **Train** | 8 000 | 264 511 | Car | 159 491 | 77 527 | 5 177 |
| | | | Truck | 4 197 | 6 262 | 1 167 |
| | | | Bus | 1 990 | 7 879 | 861 |
| **Val** | 1 000 | 34 560 | Car | 21 051 | 9 881 | 630 |
| | | | Truck | 552 | 853 | 103 |
| | | | Bus | 243 | 1 122 | 125 |
| **Test** | 2 000 | 65 322 | Car | 38 910 | 19 860 | 1 256 |
| | | | Truck | 881 | 1 409 | 263 |
| | | | Bus | 473 | 1 991 | 279 |
| **Real-world Train** | 400 | 13 753 | Car | 5 816 | 6 487 | 695 |
| | | | Truck | 86 | 204 | 57 |
| | | | Bus | 52 | 256 | 100 |
| **Real-world Test** | 200 | 5 543 | Car | 2 351 | 2 506 | 365 |
| | | | Truck | 26 | 86 | 30 |
| | | | Bus | 17 | 107 | 55 |
---
You can also **download** our HazyDet dataset from [**Baidu Netdisk**](https://pan.baidu.com/s/1KKWqTbG1oBAdlIZrTzTceQ?pwd=grok) or [**OneDrive**](https://1drv.ms/f/s!AmElF7K4aY9p83CqLdm4N-JSo9rg?e=H06ghJ).<br>
For both training and inference, the following dataset structure is required:
```
HazyDet/
├── train/
│ └── clean images/
│ └── hazy images/
│ └── lables/
├──val/
│ └── clean images/
│ └── hazy images/
│ └── lables/
├── test/
│ └── clean images/
│ └── hazy images/
│ └── lables/
├── Real-world/
│ └── train/
│ └── test/
│ └── lables/
└── README.md <-- you are here
```
**Note: Both passwords for BaiduYun and OneDrive is `grok`**.
## Leadboard and Model Zoo
All the weight files in the model zoo can be accessed on [Baidu Cloud](https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok) and [OneDrive](https:).
### Detectors
| Model | Backbone | #Params (M) | GFLOPs | mAP on<br>Synthetic Test-set | mAP on<br>Real-world Test-set | Weight |
|-------------------------|----------|-------------|--------|-----------------------------|-------------------------------|--------|
| **One Stage** | | | | | | |
| YOLOv3 | Darknet53 | 61.63 | 20.19 | 35.0 | 30.7 | [weight](https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok) |
| GFL | ResNet50 | 32.26 | 198.65 | 36.8 | 32.5 | [weight](https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok) |
| YOLOX | CSPDarkNet | 8.94 | 13.32 | 42.3 | 35.4 | [weight](https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok) |
| FCOS | ResNet50 | 32.11 | 191.48 | 45.9 | 32.7 | [weight](https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok) |
| VFNet | ResNet50 | 32.71 | 184.32 | 49.5 | 35.6 | [weight](https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok) |
| ATTS | ResNet50 | 32.12 | 195.58 | 50.4 | 36.4 | [weight](https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok) |
| DDOD | ResNet50 | 32.20 | 173.05 | 50.7 | 37.1 | [weight](https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok) |
| TOOD | ResNet50 | 32.02 | 192.51 | 51.4 | 36.7 | [weight](https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok) |
| **Two Stage** | | | | | | |
| Faster RCNN | ResNet50 | 41.35 | 201.72 | 48.7 | 33.4 | [weight](https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok) |
| Libra RCNN | ResNet50 | 41.62 | 209.92 | 49.0 | 34.5 | [weight](https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok) |
| Grid RCNN | ResNet50 | 64.46 | 317.44 | 50.5 | 35.2 | [weight](https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok) |
| Cascade RCNN | ResNet50 | 69.15 | 230.40 | <u>51.6</u> | <u>37.2</u> | [weight](https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok) |
| **End-to-End** | | | | | | |
| Conditional DETR | ResNet50 | 43.55 | 91.47 | 30.5 | 25.8 | [weight](https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok) |
| DAB DETR | ResNet50 | 43.7 | 91.02 | 31.3 | 27.2 | [weight](https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok) |
| Deform DETR | ResNet50 | 40.01 | 203.11 | 51.5 | 36.9 | [weight](https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok) |
| **DeCoDet** | | | | | | |
| **DeCoDet (Ours)** | ResNet50 | 34.62 | 225.37 | **52.0** | **38.7** | [weight](https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok) |
### Dehazing
<table>
<tr>
<td>Type</td>
<td>Method</td>
<td>PSNR</td>
<td>SSIM</td>
<td>mAP on Test-set</td>
<td></td>
<td>mAP on RDDTS</td>
<td>Weight</td>
</tr>
<tr>
<td>Baseline</td>
<td>Faster RCNN</td>
<td>-</td>
<td>-</td>
<td>39.5</td>
<td></td>
<td>21.5</td>
<td><a href="https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok">weight</a></td>
</tr>
<tr>
<td>Dehaze</td>
<td>GridDehaze</td>
<td>12.66</td>
<td>0.713</td>
<td>38.9 (-0.6)</td>
<td></td>
<td>19.6 (-1.9)</td>
<td><a href="https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok">weight</a></td>
</tr>
<tr>
<td>Dehaze</td>
<td>MixDehazeNet</td>
<td>15.52</td>
<td>0.743</td>
<td>39.9 (+0.4)</td>
<td></td>
<td>21.2 (-0.3)</td>
<td><a href="https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok">weight</a></td>
</tr>
<tr>
<td>Dehaze</td>
<td>DSANet</td>
<td>19.01</td>
<td>0.751</td>
<td>40.8 (+1.3)</td>
<td></td>
<td>22.4 (+0.9)</td>
<td><a href="https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok">weight</a></td>
</tr>
<tr>
<td>Dehaze</td>
<td>FFA</td>
<td>19.25</td>
<td>0.798</td>
<td>41.2 (+1.7)</td>
<td></td>
<td>22.0 (+0.5)</td>
<td><a href="https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok">weight</a></td>
</tr>
<tr>
<td>Dehaze</td>
<td>DehazeFormer</td>
<td>17.53</td>
<td>0.802</td>
<td>42.5 (+3.0)</td>
<td></td>
<td>21.9 (+0.4)</td>
<td><a href="https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok">weight</a></td>
</tr>
<tr>
<td>Dehaze</td>
<td>gUNet</td>
<td>19.49</td>
<td>0.822</td>
<td>42.7 (+3.2)</td>
<td></td>
<td>22.2 (+0.7)</td>
<td><a href="https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok">weight</a></td>
</tr>
<tr>
<td>Dehaze</td>
<td>C2PNet</td>
<td>21.31</td>
<td>0.832</td>
<td>42.9 (+3.4)</td>
<td></td>
<td>22.4 (+0.9)</td>
<td><a href="https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok">weight</a></td>
</tr>
<tr>
<td>Dehaze</td>
<td>DCP</td>
<td>16.98</td>
<td>0.824</td>
<td>44.0 (+4.5)</td>
<td></td>
<td>20.6 (-0.9)</td>
<td><a href="https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok">weight</a></td>
</tr>
<tr>
<td>Dehaze</td>
<td>RIDCP</td>
<td>16.15</td>
<td>0.718</td>
<td>44.8 (+5.3)</td>
<td></td>
<td>24.2 (+2.7)</td>
<td><a href="https://pan.baidu.com/s/1EEX_934Q421RkHCx53akJQ?pwd=grok">weight</a></td>
</tr>
</table>
## Citation
If you use this toolbox or benchmark in your research, please cite this project.
```bibtex
@article{feng2025HazyDet,
title={HazyDet: Open-Source Benchmark for Drone-View Object Detection with Depth-Cues in Hazy Scenes},
author={Changfeng Feng and Zhenyuan Chen and Xiang Li and Chunping Wang and Jian Yang and Ming-Ming Cheng and Yimian Dai and Qiang Fu},
year={2025},
journal={arXiv preprint arXiv:2409.19833},
}
@article{zhu2021detection,
title={Detection and tracking meet drones challenge},
author={Zhu, Pengfei and Wen, Longyin and Du, Dawei and Bian, Xiao and Fan, Heng and Hu, Qinghua and Ling, Haibin},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
volume={44},
number={11},
pages={7380--7399},
year={2021},
publisher={IEEE}
}
``` |