Scholarlink

Different spectral images contain different information, and fusing multispectral information can enhance the performance considerably of object detection. We introduce a multimodal object detection network that can effectively achieve feature-level fusion of RGB images and infrared images with lightweight network. Firstly, we utilize a CrossModality Deformable Transformer (CDT) module to capture the latent relationship between the two modalities. Additionally, we design a dual-stream network structure to fuse light images and infrared images while balancing the model parameter volume. Experimental results showed that this method attains the top results on the VEDAI, FLIR, and LLVIP datasets.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Cross-Modality Fusion Deformable Transformer for Multispectral Object Detection

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Yigan Wang,
  • Chaoying Tang,
  • Qiaohan Shi

摘要

Different spectral images contain different information, and fusing multispectral information can enhance the performance considerably of object detection. We introduce a multimodal object detection network that can effectively achieve feature-level fusion of RGB images and infrared images with lightweight network. Firstly, we utilize a CrossModality Deformable Transformer (CDT) module to capture the latent relationship between the two modalities. Additionally, we design a dual-stream network structure to fuse light images and infrared images while balancing the model parameter volume. Experimental results showed that this method attains the top results on the VEDAI, FLIR, and LLVIP datasets.