pseshadri9/ASPED

---
license: cc-by-4.0
task_categories:
- audio-classification
size_categories:
- n>1T
tags:
- pedestrian detection
---
# ASPED: An Audio Dataset for Detecting Pedestrians

This repo contains the data for the ASPED dataset, presented at ICASSP 2024.
- [Paper Link](https://arxiv.org/abs/2309.06531), [Project Homepage](https://urbanaudiosensing.github.io/ASPED.html)

- Pavan Seshadri, Chaeyeon Han, Bon-Woo Koo, Noah Posner, Suhbrajit Guhathakurta, Alexander Lerch

## Usage
This dataset contains audio and video recordings of pedestrian activity collected at various locations in and around Georgia Tech.

Labels of pedestrian counts per each second of audio/video are provided as well, calculated via a computer vision model (Mask2Former trained on msft-coco) using the video recordings.
### Access
It is recommended to use the huggingface_hub library to download the dataset from this location. [Info on downloading with huggingface_hub](https://huggingface.co/docs/huggingface_hub/guides/download).

Downloading the entire dataset can be done with the following code:
```
from huggingface_hub import snapshot_download
snapshot_download(repo_id="pseshadri9/ASPED", repo_type="dataset")
```
Alternatively if you would like to download only the audio or video, pass the ignore_patterns flag to snapshot_download to avoid downloading the entire set.

**Audio Only**
```
from huggingface_hub import snapshot_download
snapshot_download(repo_id="pseshadri9/ASPED", repo_type="dataset", ignore_patterns="*.mp4")
```

**Video Only**

```
from huggingface_hub import snapshot_download
snapshot_download(repo_id="pseshadri9/ASPED", repo_type="dataset", ignore_patterns="*.flac")
```

## Citation
```
@inproceedings{Seshadri24,
title={ASPED: An Audio Dataset for Detecting Pedestrians},
author={Seshadri, Pavan and Han, Chaeyeon and Koo, Bon-Woo and Posner, Noah and Guhathakurta, Suhbrajit and Lerch, Alexander},
booktitle={Proc. of ICASSP 2024},
pages={1--5},
year={2024},
organization={IEEE}
}
```

原始信息汇总

ASPED: An Audio Dataset for Detecting Pedestrians

概述

ASPED 数据集包含在佐治亚理工学院及其周边地区收集的行人活动的音频和视频记录。每秒音频/视频的行人数量标签也已提供，这些标签是通过使用计算机视觉模型（Mask2Former 在 msft-coco 上训练）处理视频记录计算得出的。

访问

建议使用 huggingface_hub 库从指定位置下载数据集。下载整个数据集的代码如下： python from huggingface_hub import snapshot_download snapshot_download(repo_id="pseshadri9/ASPED", repo_type="dataset")

如果只需要下载音频或视频，可以使用 ignore_patterns 标志避免下载整个数据集。

仅下载音频： python from huggingface_hub import snapshot_download snapshot_download(repo_id="pseshadri9/ASPED", repo_type="dataset", ignore_patterns="*.mp4")

仅下载视频： python from huggingface_hub import snapshot_download snapshot_download(repo_id="pseshadri9/ASPED", repo_type="dataset", ignore_patterns="*.flac")

引用

plaintext @inproceedings{Seshadri24, title={ASPED: An Audio Dataset for Detecting Pedestrians}, author={Seshadri, Pavan and Han, Chaeyeon and Koo, Bon-Woo and Posner, Noah and Guhathakurta, Suhbrajit and Lerch, Alexander}, booktitle={Proc. of ICASSP 2024}, pages={1--5}, year={2024}, organization={IEEE} }