ReFusion is a novel approach that leverages meta-learning to optimize fusion loss for various tasks, demonstrating superior performance in infrared-visible, medical, multi-focus, and multi-exposure image fusion. The framework consists of three key components: a fusion module, a loss proposal module, and a source reconstruction module. Through alternating updates of these modules, ReFusion achieves high-quality fusion results by preserving information from source images and adapting the fusion loss dynamically.
The content discusses the challenges in traditional image fusion methods due to the lack of definitive ground truth and distance measurement. It highlights the importance of adaptive loss functions tailored to specific scenarios and tasks. The learnable loss function proposed by ReFusion assigns pixel-wise preferences to source images based on intensity and gradient aspects, dynamically adjusting during training.
Extensive experiments showcase ReFusion's effectiveness in various image fusion tasks. The framework's innovative approach to learning optimal fusion loss through meta-learning sets it apart from existing methods.
In un'altra lingua
dal contenuto originale
arxiv.org
Approfondimenti chiave tratti da
by Haowen Bai,Z... alle arxiv.org 03-12-2024
https://arxiv.org/pdf/2312.07943.pdfDomande più approfondite