# AG-VPReID 2025: The 2nd Aerial-Ground Person ReID — Data

来源：https://www.kaggle.com/competitions/agvpreid25/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

# Author Kit

**Download Method Description Template**

[📥 Download Author Kit](https://storage.googleapis.com/kaggle-competitions-data/kaggle-v2/94037/11295963/upload/AG-VPReID2025_authorkit.zip?GoogleAccessId=web-data@kaggle-161607.iam.gserviceaccount.com&Expires=1749876196&Signature=NmhnrCET1AvLeFAnYaTYj8vMhy%2BmvEp958wq8hASPw1RiIIdvIFpSiFgs%2BKNbTu6asCYuTSgBfUFIRBF%2BZnCv9zvPBIQw1N%2BFtSLcYmf47%2BLx7fjw6%2FrN09n14Ts3rKT181W2Izuw8%2BVOkNCjsdBlqOTWxET0AADdsm2oKkDtTCKqun8A62Zhymz9eNklGvPaT%2B2FoyIKmV%2BchsE%2FHcPSG%2B4HoiC40O58OJtnafqjrKUtGldlyoOO1gfJN4QVS%2FihtP5wQkNpml%2B5P%2FTeCfkj2mNKUwt84PZeBQ8JmqjZIhcK9oBQdfP2AMBjd3x%2BxGPB%2F2EDMACkQPzt%2BZLFyMlGw%3D%3D&response-content-disposition=attachment%3B+filename%3DAG-VPReID2025_authorkit.zip)

Alternatively, please follow **Note about downloading data** in the below section to download the template.

**Important:**
- Before downloading, please read the [Privacy and Usage Policy](www.kaggle.com/competitions/agvpreid25/overview/privacy)
- For competition integrity, IDs have been anonymized. Query IDs **DO NOT** directly correspond to Gallery IDs in either testing case (this masking also applies to the attribute files).

**Note about downloading data:** The "**Download All**" button may **not function** properly. Instead, navigate to the specific folder you need in the Data Explorer, select it, and then click the download icon to retrieve the files.

## Training Dataset
| Identities | Tracklets | Frames |
|:----------:|:---------:|:------:|
| 689 | 5,317 | ~1.47M |

## Testing Dataset

### Testing Case 1: Aerial → Ground
| Set | Identities | Tracklets | Frames |
|:----|:----------:|:---------:|:------:|
| Query | 645 | 3,023 | 424,532 |
| Gallery | 645 | 2,750 | 1,126,213 |

### Testing Case 2: Ground → Aerial
| Set | Identities | Tracklets | Frames |
|:----|:----------:|:---------:|:------:|
| Query | 645 | 2,750 | 1,126,213 |
| Gallery | 645 | 3,023 | 424,532 |
| Gallery Distractor | 1,693 | 2,417 | 687,988 |

## Data Structure
- Each video tracklet: `{ID}/{Tracklet}/{Image/Frame}`
- Frame format e.g P81T2310171A2C2E152K7F7528 breaks down as:
- P81: Person identifier for collection day only (not permanent ID)
- T231017: Collection date (Oct 17, 2023)
- 1A2: Altitude indicator (A2 = 80m, A3 = 120m)
- C2: Camera type (C0-C6 representing CCTVx2/GoProx2/UAVx2)
- E152: Event identifier (for annotation purpose)
- K7: Key point reference (for annotation purpose)
- F7528: Frame number from video

## Files Description
- `train/` - Training tracklets organized by person ID
- `attributes/` - 15 soft-biometric attributes (CSV format) for train and testing cases

### Testing Case 1: Aerial to Ground
- `case1_aerial_to_ground_query/` - Query tracklets
- `case1_aerial_to_ground_gallery/` - Gallery tracklets

### Testing Case 2: Ground to Aerial
- `case2_ground_to_aerial_query/` - Query tracklets
- `case2_ground_to_aerial_gallery/` - Gallery tracklets

### Submission Files
- `sample_submission.csv` - Example submission format
- `submission.csv` - Template for result submission
- `AG-VPReID2025_authorkit/` - Template for method description document

**Note:** In `submission.csv`:
- Row IDs 1-3023: Case 1 (Aerial→Ground) queries
- Row IDs 3024-5774: Case 2 (Ground→Aerial) queries

## 📖 Citation

If you use our dataset in your research, please cite our work:

`Nguyen, H., Nguyen, K., Pemasiri, A., Liu, F., Sridharan, S., & Fookes, C. (2025). AG-VPReID: A challenging large-scale benchmark for aerial-ground video-based person re-identification. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)`