shiniagarwal/drone-detections
收藏Hugging Face2026-03-23 更新2026-03-29 收录
下载链接:
https://hf-mirror.com/datasets/shiniagarwal/drone-detections
下载链接
链接失效反馈官方服务:
资源简介:
---
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
dataset_info:
features:
- name: image
dtype: image
- name: bbox
list: float32
- name: confidence
dtype: float32
- name: video
dtype: string
- name: frame_idx
dtype: int32
splits:
- name: train
num_bytes: 23278829.0
num_examples: 124
download_size: 23281682
dataset_size: 23278829.0
---
# Drone Detection Frames
Detection frames extracted from two UAV drone videos as part of Assignment 3 (AI Spring 2026 — University of the Aegean).
## Dataset Description
Frames were extracted from two YouTube videos containing drone footage. A YOLOv8n model fine-tuned on the [Drone Detection dataset (v6)](https://universe.roboflow.com/drone-detection-g4d3g/drone-detection-a1tsf/dataset/6) was used to detect drones in every frame. Only frames with at least one detection were saved.
Out of 3599 total detection frames across both videos, every 10th frame was uploaded (124 frames total) to keep the dataset size manageable.
## Columns
| Column | Type | Description |
|--------|------|-------------|
| `image` | Image | The video frame (resized to 640×360) |
| `bbox` | List[float] | Bounding box of the detected drone [x1, y1, x2, y2] |
| `confidence` | float | Detection confidence score from YOLOv8 |
| `video` | string | Which video the frame came from (drone_video_1 or drone_video_2) |
| `frame_idx` | int | Frame number in the original video |
## Source Videos
- [Video 1](https://youtu.be/YNV7JmNZA5M) — tracked output
- [Video 2](https://youtu.be/9MWnNpHjkoY) — tracked output
## Model
YOLOv8n fine-tuned for 30 epochs on the Roboflow Drone Detection dataset (v6). Confidence threshold of 0.3 used at inference.
## License
CC BY 4.0
配置项:
- 配置名称:default
数据文件:
- 拆分集:训练集
路径:data/train-*
数据集信息:
特征:
- 名称:image
数据类型:图像
- 名称:bbox
类型:浮点型列表(float32)
- 名称:confidence
数据类型:单精度浮点型(float32)
- 名称:video
数据类型:字符串
- 名称:frame_idx
数据类型:32位整型(int32)
拆分集信息:
- 名称:训练集
字节大小:23278829.0
样本数量:124
下载大小:23281682
数据集总大小:23278829.0
# 无人机检测帧数据集
本数据集为爱琴海大学2026年春季人工智能课程第三次作业的组成部分,包含从两段无人机(UAV, Unmanned Aerial Vehicle)视频中提取的检测帧。
## 数据集说明
本数据集的帧提取自两段包含无人机航拍素材的YouTube视频。我们使用在[无人机检测数据集(v6)](https://universe.roboflow.com/drone-detection-g4d3g/drone-detection-a1tsf/dataset/6)上微调完成的YOLOv8n模型,对每一帧进行无人机检测,仅保留包含至少一个检测结果的帧。
两段视频总计生成3599张有效检测帧,为控制数据集规模,我们仅选取每10帧中的1张进行上传,最终共包含124个样本。
## 数据列说明
| 列名 | 数据类型 | 描述 |
|--------|------|-------------|
| `image` | 图像 | 视频帧(已统一调整至640×360分辨率) |
| `bbox` | 浮点型列表 | 检测到的无人机边界框,格式为[x1, y1, x2, y2] |
| `confidence` | 单精度浮点型 | YOLOv8n模型输出的检测置信度得分 |
| `video` | 字符串 | 该帧所属的视频来源,取值为drone_video_1或drone_video_2 |
| `frame_idx` | 整型 | 该帧在原始视频中的帧序号 |
## 源视频
- [视频1](https://youtu.be/YNV7JmNZA5M) — 追踪输出结果
- [视频2](https://youtu.be/9MWnNpHjkoY) — 追踪输出结果
## 检测模型
本数据集使用在Roboflow无人机检测数据集(v6)上微调30个训练轮次的YOLOv8n模型,推理阶段置信度阈值设置为0.3。
## 许可证
CC BY 4.0
提供机构:
shiniagarwal



