UAVDB: Trajectory-Guided Adaptable Bounding Boxes for UAV Detection

This repository introduces the Patch Intensity Convergence (PIC) technique, a trajectory-guided method for generating high-fidelity UAV detection bounding boxes without manual labeling. It also serves as the foundation for UAVDB, a dedicated database for UAV detection. More details are presented in the paper.

1. Environment Setup

Hardware Information

Laptop

• CPU: Intel® Core™ i7-12650H
• GPU: NVIDIA GeForce RTX 4050 Laptop GPU (6GB)
• RAM: 23734MiB

HPC

• GPU: Spartan gpu-a100 (80GB)

Folder Structure

UAVDB/
├── assets/
├── logfile/
├── src/PIC.py
├── data/
│   ├── videos/d1c0.txt, d1c1.txt, ...
│   └── detections/d1c0.mp4, d1c1.mp4, ...
├── datasets/
│   ├── train/d1c0_0.jpg, d1c0_0.txt, ...
│   ├── val/d1c1_0.jpg, d1c1_0.txt, ...
│   └── test/d1c3_0.jpg, d1c3_0.txt, ...
└── weights/
    ├── yolov8n.pt
    ├── yolov8s.pt
    ├── yolov9t.pt
    ├── yolov9s.pt
    ├── yolo10n.pt
    ├── yolo10s.pt
    ├── yolo11n.pt
    ├── yolo11s.pt
    ├── yolov12n.pt
    └── yolov12s.pt

Conda Environments

# PIC
$ conda create -n pic python=3.10 -y
$ conda activate pic
$ conda install -c conda-forge cupy
$ pip install scipy opencv-python opencv-contrib-python matplotlib numba


# YOLOv8 & YOLO11
$ conda create -n ultralytics python=3.10 -y
$ conda activate ultralytics
$ pip install ultralytics seaborn
$ git clone https://github.com/ultralytics/ultralytics.git
$ cd ultralytics/

# download pretrained weights
$ wget https://github.com/ultralytics/assets/releases/download/v8.3.0/yolo11n.pt
$ wget https://github.com/ultralytics/assets/releases/download/v8.3.0/yolo11s.pt
$ wget https://github.com/ultralytics/assets/releases/download/v8.3.0/yolo11m.pt
$ wget https://github.com/ultralytics/assets/releases/download/v8.3.0/yolo11l.pt
$ wget https://github.com/ultralytics/assets/releases/download/v8.3.0/yolo11x.pt


# YOLOv9
$ conda create -n yolov9 python=3.10 -y
$ conda activate yolov9
$ git clone https://github.com/WongKinYiu/yolov9.git
$ cd yolov9/
$ pip install -r requirements.txt
$ pip install Pillow==9.5.0

# download pretrained weights
$ wget https://github.com/WongKinYiu/yolov9/releases/download/v0.1/yolov9-t-converted.pt
$ wget https://github.com/WongKinYiu/yolov9/releases/download/v0.1/yolov9-s-converted.pt
$ wget https://github.com/WongKinYiu/yolov9/releases/download/v0.1/yolov9-m-converted.pt
$ wget https://github.com/WongKinYiu/yolov9/releases/download/v0.1/yolov9-c-converted.pt
$ wget https://github.com/WongKinYiu/yolov9/releases/download/v0.1/yolov9-e-converted.pt


# YOLOv10
# setup
$ conda create -n yolov10 python=3.10 -y
$ conda activate yolov10
$ git clone https://github.com/THU-MIG/yolov10.git
$ cd yolov10/
$ pip install -r requirements.txt
$ pip install -e .

# download pretrained weights
$ wget https://github.com/THU-MIG/yolov10/releases/download/v1.1/yolov10n.pt
$ wget https://github.com/THU-MIG/yolov10/releases/download/v1.1/yolov10s.pt
$ wget https://github.com/THU-MIG/yolov10/releases/download/v1.1/yolov10m.pt
$ wget https://github.com/THU-MIG/yolov10/releases/download/v1.1/yolov10b.pt
$ wget https://github.com/THU-MIG/yolov10/releases/download/v1.1/yolov10l.pt
$ wget https://github.com/THU-MIG/yolov10/releases/download/v1.1/yolov10x.pt


# YOLOv12
# setup
$ conda create -n yolov12 python=3.11 -y
$ conda activate yolov12
$ git clone https://github.com/sunsmarterjie/yolov12.git
$ cd yolov12/
$ wget https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.3/flash_attn-2.7.3+cu11torch2.2cxx11abiFALSE-cp311-cp311-linux_x86_64.whl
$ pip install -r requirements.txt
$ pip install -e .

# download pretrained weights
$ wget https://github.com/sunsmarterjie/yolov12/releases/download/v1.0/yolov12n.pt
$ wget https://github.com/sunsmarterjie/yolov12/releases/download/v1.0/yolov12s.pt
$ wget https://github.com/sunsmarterjie/yolov12/releases/download/v1.0/yolov12m.pt
$ wget https://github.com/sunsmarterjie/yolov12/releases/download/v1.0/yolov12l.pt
$ wget https://github.com/sunsmarterjie/yolov12/releases/download/v1.0/yolov12x.pt

2. Dataset Description

Multi-view drone tracking datasets

Summary of dataset characteristics in Multi-view drone tracking datasets. The table displays the number of frames and resolution for each camera across different datasets. Each cell lists the number of frames followed by the resolution in pixels.

Camera $$\backslash$$ Dataset	1	2	3	4	5
0	5334 / 1920×1080	4377 / 1920×1080	33875 / 1920×1080	31075 / 1920×1080	20970 / 1920×1080
1	4941 / 1920×1080	4749 / 1920×1080	19960 / 1920×1080	15409 / 1920×1080	28047 / 1920×1080
2	8016 / 1920×1080	8688 / 1920×1080	17166 / 3840×2160	15678 / 1920×1080	31860 / 2704×2028
3	4080 / 1920×1080	4332 / 1920×1080	14196 / 1440×1080	10933 / 3840×2160	31992 / 1920×1080
4	--	--	18900 / 1920×1080	17640 / 1920×1080	21523 / 2288×1080
5	--	--	28080 / 1920×1080	32016 / 1920×1080	17550 / 1920×1080
6	--	--	--	11292 / 1440×1080	--

UAVDB

Overview of UAVDB constructed using the PIC approach, showing the distribution of 10,763 training, 2,720 validation, and 4,578 test images (18,061 total) across datasets and camera configurations.

Camera $$\backslash$$ Dataset	1	2	3	4
0	train / 291	test / 237	train / 3190	train / 2355
1	valid / 303	train / 343	train / 841	test / 416
2	train / 394	train / 809	valid / 1067	train / 701
3	test / 348	valid / 426	train / 638	train / 727
4	--	--	test / 1253	valid / 924
5	--	--	train / 1303	train / 1110
6	--	--	--	test / 385

3. Demonstration

Patch Intensity Convergence (PIC)

Validation Performance

Inference Results

4. UAVDB Benchmark Results

Model	image size	batch size	$$AP^{val}_{50}$$	$$AP^{val}_{50-95}$$	$$AP^{test}_{50}$$	$$AP^{test}_{50-95}$$
yolov8n.pt	640	32	0.829	0.522	0.789	0.450
yolov8s.pt	640	32	0.814	0.545	0.796	0.450
yolov9t.pt	640	32	0.839	0.501	0.848	0.508
yolov9s.pt	640	32	0.819	0.517	0.834	0.484
yolo10n.pt	640	32	0.764	0.492	0.731	0.417
yolo10s.pt	640	32	0.817	0.530	0.823	0.516
yolo11n.pt	640	32	0.847	0.527	0.856	0.539
yolo11s.pt	640	32	0.826	0.553	0.885	0.578
yolov12n.pt	640	32	0.857	0.544	0.848	0.531
yolov12s.pt	640	32	0.869	0.566	0.882	0.565

Citation

If you find this project helpful for your research or applications, we would appreciate it if you could give it a star and cite the paper.

@article{chen2024uavdb,
  title={UAVDB: Trajectory-Guided Adaptable Bounding Boxes for UAV Detection},
  author={Chen, Yu-Hsi},
  journal={arXiv preprint arXiv:2409.06490},
  year={2024}
}

Acknowledgment

The data and evaluation codes based on the Multi-view Drone Tracking Datasets, YOLOv8, YOLO11, YOLOv9, YOLO10, and YOLOv12 and greatly appreciate their excellent works.