README.md

---
license: cc-by-nc-nd-4.0
task_categories:
- object-detection
- image-classification
- video-classification
- feature-extraction
language:
- en
tags:
- cross-view-perception
- urban-traffic
- traffic-dataset
- aerial-ground
- drone-vision
- monocular-bev
- ntersection-monitoring
- computer-vision
pretty_name: Cross-View Urban Traffic Dataset
size_categories:
- 1K<n<10K
configs:
- config_name: default
  data_files:
  - split: train
    path:
    - scene_03_04/raw/gt_pairs.csv
---
# Cross-View Urban Traffic Subset Dataset

This repository provides a **representative sample** of the **Cross-View Urban Traffic Dataset (CVUTD)** for reviewer inspection and qualitative verification.

The subset is intended to let reviewers:
- inspect the raw and annotated data format,
- verify annotation quality,
- understand the cross-view correspondence structure,
- and assess the benchmark design without downloading the full dataset.

The **full dataset** is hosted separately and is available at submission time:
- **Full dataset:** https://huggingface.co/datasets/prakharbh/CrossViewUrbanTrafficDataset

---

## Dataset Summary

The Cross-View Urban Traffic Dataset is a benchmark for **cross-view urban traffic perception** built from synchronized:
- **ego-centric bicycle-mounted videos**, and
- **aerial drone videos**

recorded over real urban intersections.

The benchmark supports two linked tasks:
1. **Cross-view identity matching** between street-view and drone-view object tracks
2. **Ego-to-BEV prediction** using aerial supervision

This subset repository contains a **small but representative sample scene** from the full dataset.

---

## What This Subset Contains

This subset contains one representative scene with synchronized cross-view traffic data and benchmark annotations.

Typical files included are:
- `street.mp4` or `street_annotated.mp4`
- `drone.mp4` or `drone_annotated.mp4`
- `street_annotations.csv`
- `drone_annotations.csv`
- `gt_pairs.csv`

Depending on the uploaded version, additional processed files may also be present:
- `frame_matches.csv`
- `track_mapping.csv`
- `coord_align.csv`

These files are sufficient for reviewers to inspect:
- scene structure,
- tracked detections in both views,
- manually verified cross-view ground-truth correspondences,
- and the benchmark’s tabular annotation format.

---

## How This Sample Was Created

This subset was created by selecting **one complete scene** from the full benchmark release and publishing its associated benchmark files as a lightweight, reviewable sample.

The sample was chosen to be **representative** of the full dataset in the following ways:
- it contains synchronized **ego-view and drone-view** observations of the same urban intersection scene,
- it includes tracked traffic participants in both views,
- it contains manually verified **cross-view identity correspondences** in `gt_pairs.csv`,
- and it reflects the same annotation and preprocessing format used throughout the full dataset.

This subset is **not a different dataset** and does not use a different processing pipeline. It is simply a reduced release of the same benchmark format used in the full dataset.

---

## File Format Description

### `street_annotations.csv`
Street-view tracked detections, typically including:
- frame index
- track ID
- class label
- bounding box coordinates

### `drone_annotations.csv`
Drone-view tracked detections, typically including:
- frame index
- track ID
- class label
- bounding box coordinates
- metric world coordinates (when available)

### `gt_pairs.csv`
Human-verified cross-view ground-truth correspondences:
- `scene_id`
- `street_track_id`
- `drone_track_id`
- `class_name`

This file defines the benchmark identity supervision linking street-view and drone-view tracks.

---

## Intended Purpose of the Subset

This subset is provided to support:
- reviewer inspection,
- format verification,
- benchmark transparency,
- and lightweight qualitative inspection of the released data.

It is **not intended** to replace the full dataset for training or large-scale benchmarking.

For full experiments, use the complete dataset release:
- https://huggingface.co/datasets/prakharbh/CrossViewUrbanTrafficDataset

---

## Relationship to the Full Dataset

This repository is a **subset release** of the full Cross-View Urban Traffic Dataset.

The full dataset contains:
- multiple urban intersections,
- substantially more tracked traffic participants,
- more cross-view correspondences,
- and the full benchmark scope used in the paper.

This subset exists only to provide a compact and accessible sample for inspection and review.

---

## Privacy and Anonymization

The data were recorded in real urban traffic environments. To support responsible release:
- human faces are anonymized,
- vehicle license plates are anonymized,
- and the subset is intended only for research and benchmark inspection.

The dataset is **not intended** for surveillance, face recognition, license-plate recognition, or person re-identification.

---

## License

This subset is released under the same license as the full dataset:

**CC-BY-NC-ND 4.0**

Please refer to the full dataset page for release details and usage conditions.

---

## Code and Benchmark Pipeline

Code, preprocessing scripts, annotation tools, baseline implementations, and evaluation scripts are available at:

- **GitHub repository:** https://github.com/oth-aifiud/Cross-View-Urban-Traffic-Dataset

For whole-dataset processing and evaluation across multiple scenes, use the batch scripts in the `batch_scripts/` directory of the code repository.

---

## Citation

If you use this dataset or subset in research, please cite the associated paper and dataset release.

```bibtex
@misc{crossviewurbantrafficdataset,
  title={Cross-View Urban Traffic Dataset},
  author={Prakhar Bhardwaj and collaborators},
  year={2026},
  howpublished={Hugging Face dataset repository},
  url={https://huggingface.co/datasets/prakharbh/CrossViewUrbanTrafficDataset}
}