RGB-Th-Bench introduces the first benchmark designed to evaluate how well Vision-Language Models (VLMs) can understand and interpret thermal imagery alongside standard RGB images.

• Creates a comprehensive evaluation framework specifically for thermal-visual understanding
• Addresses a critical gap in existing benchmarks that primarily focus on RGB images
• Enables rigorous testing of AI models for security applications including surveillance and detection in low-visibility conditions
• Provides a foundation for developing more robust multimodal systems capable of operating across visual spectrums

This research significantly advances security applications by improving AI capabilities in thermal imaging - essential for 24/7 surveillance, perimeter protection, and human detection in challenging environments where traditional RGB cameras fail.

RGB-Th-Bench: A Dense benchmark for Visual-Thermal Understanding of Vision Language Models