ATR-UMMIR (Unmanned Multimodal Image Registration) is a large-scale dataset for multimodal image registration and matching in aerial scenarios. It focuses on aligning visible and infrared image pairs captured by UAVs under diverse real-world conditions. This dataset is designed to advance research in multimodal image alignment, fusion-based detection, and condition-aware visual understanding.

• Modalities: Aligned visible–infrared image pairs
• Scene Count: 15,000+ unique locations
• Total Images: 60,000+ (30k visible, 30k infrared)
• Resolution: 640×512 pixels
• Annotations:
  • Coarse-level manual alignment
  • Fine-level keypoints (for subset)
  • Detailed condition labels (see below)

To reflect real-world complexity, each image pair is annotated across six condition attributes:

• Altitude: 80m–300m (majority in 100–120m)
• Camera Angle: 0° (nadir) to 75° (oblique), majority at 30°–45°
• Shooting Time: Day, night, dawn, morning, afternoon
• Weather: Sunny, cloudy, rainy, after-rain, foggy
• Illumination: Night, twilight, dim, normal, overexposed
• Scenario: 11 types including urban, suburban, village, factory, road, school, etc.

This rich condition diversity enables robust evaluation of multimodal models under dynamic imaging environments.

[⏬ Dataset Download Link (Link)]

Please fill in the download address or contact us for access.

• Multimodal image registration and alignment
• Condition-aware image matching
• Cross-modality fusion and detection
• UAV-based remote sensing tasks

If you use ATR-UMMIR in your research, please cite:

@misc{ATRUMMIR2025,
  title={ATR-UMMIR: A Multimodal UAV Image Matching Dataset under Diverse Conditions},
  author={Your Name and Others},
  year={2025},
  howpublished={\url{https://github.com/yourname/ATR-UMMIR}},
}