RGBT-Ground: A New Benchmark for Robust Visual Grounding in Real-World Scenarios
Published:Dec 31, 2025 02:01
•1 min read
•ArXiv
Analysis
This paper introduces a new benchmark, RGBT-Ground, specifically designed to address the limitations of existing visual grounding benchmarks in complex, real-world scenarios. The focus on RGB and Thermal Infrared (TIR) image pairs, along with detailed annotations, allows for a more comprehensive evaluation of model robustness under challenging conditions like varying illumination and weather. The development of a unified framework and the RGBT-VGNet baseline further contribute to advancing research in this area.
Key Takeaways
- •Introduces RGBT-Ground, a new benchmark for visual grounding in complex real-world scenarios.
- •Utilizes RGB and Thermal Infrared (TIR) image pairs for more robust evaluation.
- •Provides a unified visual grounding framework and a baseline model (RGBT-VGNet).
- •Addresses limitations of existing benchmarks in terms of scene diversity and real-world conditions.
Reference
“RGBT-Ground, the first large-scale visual grounding benchmark built for complex real-world scenarios.”