VRAI Vehicle Re-identification Dataset

Dataset Overview

Name: VRAI Vehicle Re‑identification Dataset

Purpose: Supports vehicle re‑identification research using aerial imagery.

Background: The rapid growth of UAVs has spurred interest in UAV‑based visual applications. Yet, vehicle re‑ID research using UAV data remains limited due to the lack of publicly available datasets, which require extensive UAV flights, video capture, and manual annotation.

Dataset Splits

• Training set: 66,113 images, 6,302 IDs.
• Test set: 71,500 images, 6,720 IDs.
• Test‑dev set: 20 % of test images sampled for development.

Annotation Details

Training Set

• Image naming: ID_Cam_Frame.jpg (e.g., 0000000X_000Y_0000000Z.jpg).
• Annotation file: train_annotation.pkl containing image filenames, color labels, type labels, vehicle attribute labels, and bounding boxes for vehicle parts.

Test Set

• Image naming: RandomString_Cams.jpg (e.g., 00AV11D2_C1.jpg).
• Annotation file: test_annotation.pkl with similar fields plus gallery and query ordering.

Test‑dev Set

• Same naming convention and annotation structure as the test set.

Evaluation Metrics

• Metrics: Mean Average Precision (mAP) and Cumulative Matching Curve (CMC).
• Challenge platform: EvalAI.

Dataset Download

• Google Drive: Link
• Baidu Cloud: Link (extraction code: his6)

Usage Statements

• Commercial use: Prohibited.
• Citation requirement: Must cite relevant publications when using the dataset.