# VReID-XFD — Data

来源：https://www.kaggle.com/competitions/detreidxv1/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

# 📘 Dataset Description — DetReIDX

The **DetReIDX** dataset is a *multi-session, multi-view* benchmark for **video-based person re-identification** across **aerial (UAV)** and **ground/indoor** cameras.  
It was collected at **seven universities** spanning **Portugal, Turkey, India, and Angola**, with each identity recorded on **different days** (S1, S2) to introduce *natural variation in clothing, lighting, and location*.  
All identities are consistently labeled across sessions and annotated with **16 soft-biometric attributes**.

---

## 🧰 Author Kit

**Download Author Kit**  
Method description template (PDF/Markdown) and checklist for leaderboard reports.

> *Alternatively:* see the **Note about downloading data** below to fetch the template from the **Data** tab.

**Important**
- Please read the **Privacy & Usage Policy** before downloading.  
- For competition integrity, IDs are **anonymized**.  
  *Query IDs do not directly map to gallery IDs* in any test case (masking also applies to attribute files).
- For zips passwords, send an email to : kailash.hambarde@ubi.pt or kailas.srt@gmail.com

---
## 💾 Note about Downloading Data

Due to the large data volume (**>100 GB**), the dataset is hosted externally. You can download each part using the links below:

- **Training Set:** [train.zip](https://www.it.ubi.pt/DetReIDX_dataset/VReID-XFD/train.zip)
- **Test Set A→A:** [case1_aerial_to_ground.zip](https://www.it.ubi.pt/DetReIDX_dataset/VReID-XFD/case1_aerial_to_ground.zip)
- **Test Set A→G:** [case2_ground_to_aerial.zip](https://www.it.ubi.pt/DetReIDX_dataset/VReID-XFD/case2_ground_to_aerial.zip)
- **Test Set G→A:** [case3_aerial_to_aerial.zip](https://www.it.ubi.pt/DetReIDX_dataset/VReID-XFD/case3_aerial_to_aerial.zip)

---

## 📊 Full Dataset Statistics

| Subset | Identities | Tracklets | Frames (approx.) |
|:--|--:|--:|--:|
| **Full Dataset** | 371 | 22,572 | 23,641,438 |
| **Training Dataset** | 215 | 5,451 | 5,683,129 |

---

## 🧪 Testing Dataset

### Case 1: **Aerial → Ground (A→G)**

| Set | Identities | Tracklets | Frames |
|:---|---:|---:|---:|
| **Query** | 156 | 5,682 | 5,972,679 |
| **Gallery** | 89 | 89 | 56,987 |

---

### Case 2: **Ground → Aerial (G→A)**

| Set | Identities | Tracklets | Frames |
|:---|---:|---:|---:|
| **Query** | 89 | 89 | 56,987 |
| **Gallery** | 156 | 5,682 | 5,972,679 |

---

### Case 3: **Aerial → Aerial (A→A)**

| Set | Identities | Tracklets | Frames |
|:---|---:|---:|---:|
| **Query** | 155 | 2,789 | 2,784,461 |
| **Gallery** | 155 | 2,790 | 3,114,516 |

---

## 📁 Dataset Folder Structure

```python
VReID-XFD/
├── train/
│   ├── 0004/                             # Person ID
│   │   ├── 0004_0001/                    # Tracklet ID
│   │   │   ├── 0004_000_002_002_000_000_0312.jpg
│   │   └── 0004_0002/
│   │       ├── 0004_23_10_2024_09_18_5.8_10_30_3_0_1_1_0_-1_-1_6_0_0_6_0_178_1_0005.jpg
│   ├── ...
│
├── case1_aerial_to_ground/
│   ├── query/                            # Aerial images
│   └── gallery/                          # Ground images
│
├── case2_ground_to_aerial/
│   ├── query/                            # Ground images
│   └── gallery/                          # Aerial images
│
└── case3_aerial_to_aerial/
    ├── query/                            # Aerial images (first half)
    └── gallery/                          # Aerial images (second half)