A Data Augmentation Method Based on Multi-Modal Image Fusion for Detection and Segmentation

In the field of computer vision, effective data augmentation plays a crucial role in enhancing the robustness and generalization capability of visual models. This paper proposes a novel data augmentation method based on multimodal image fusion. Unlike traditional augmentation approaches, the proposed method focuses on synthesizing the fused samples that contain complementary scene characteristics from different modalities while actively suppressing useless and redundant information. To evaluate the effectiveness of our method, the experiments were conducted in the contexts of both object detection and semantic segmentation. The experimental results demonstrate that our method can significantly improve the accuracy of visual models than original samples.