Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
The paper introduces a hybrid image fusion framework that combines a learnable U-Net for generating dynamic guidance maps with a fixed Laplacian pyramid fusion kernel. This approach decouples policy learning from pixel synthesis, enabling efficient full-resolution training and eliminating the train-inference gap. The proposed method achieves state-of-the-art-comparable performance with only one minute of training on an RTX 4090 and demonstrates strong zero-shot generalization across various image fusion tasks.
Achieve state-of-the-art image fusion in just one minute of training, bridging the gap between slow deep learning methods and fast but less adaptable traditional techniques.
Image fusion seeks to integrate complementary information from multiple sources into a single, superior image. While traditional methods are fast, they lack adaptability and performance. Conversely, deep learning approaches achieve state-of-the-art (SOTA) results but suffer from critical inefficiencies: their reliance on slow, resource-intensive, patch-based training introduces a significant gap with full-resolution inference. We propose a novel hybrid framework that resolves this trade-off. Our method utilizes a learnable U-Net to generate a dynamic guidance map that directs a classic, fixed Laplacian pyramid fusion kernel. This decoupling of policy learning from pixel synthesis enables remarkably efficient full-resolution training, eliminating the train-inference gap. Consequently, our model achieves SOTA-comparable performance in about one minute on a RTX 4090 or two minutes on a consumer laptop GPU from scratch without any external model and demonstrates powerful zero-shot generalization across diverse tasks, from infrared-visible to medical imaging. By design, the fused output is linearly constructed solely from source information, ensuring high faithfulness for critical applications. The codes are available at https://github.com/Zirconium233/HybridFusion