<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[DTGshaodow的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/DTGshaodow</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; DTGshaodow]]></copyright><item><title><![CDATA[《Detection Drives an End-to-End Fusion of Infraredand Visible Images Based on Diffusion Models》论文分享]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/168453852</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/168453852</guid><author>DTGshaodow</author><pubDate>Tue, 03 Nov 2026 18:04:06 +0800</pubDate><description><![CDATA[原文链接：https://doi.org/10.1109/tip.2026.3671618author={Huang, Kexin and Zhang, Zhiyuan and Shi, Chaohua and Luo, Lan and Shi, Junpeng and Liu, Yongxiang}Abstract红外与可见光图像融合方法已取得良好效果，但现有方法要么通过独立的融合过程牺牲了下游检测性能，要么在融合与检测模型需联合训练的情况下，牺牲了计算效率和灵活性。为解决这些问题，我们提出了一种基于扩散]]></description><category></category></item><item><title><![CDATA[《DDFD: Diffusion-Based Denoising Fusion for Object Detection inInfrared-Visible Images》论文分享（侵删）]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/168453665</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/168453665</guid><author>DTGshaodow</author><pubDate>Mon, 26 Oct 2026 10:43:00 +0800</pubDate><description><![CDATA[原文链接：https://doi.org/10.1145/3746027.3755183author={Dang, Min and Liu, Gang and Zhao, Jingqi and Kong, Adams Wai-Kin and Luo, Nan and Wang, Di}Abstract红外-可见光图像融合目标检测(IVIF-OD)的目的是利用两种模式的互补信息来合成具有更丰富信息的新图像来服务于目标检测。现有的大多数工作都集中在如何更好地融合像素级细节，而忽略了检测所需的与目标相关的信息，并]]></description><category></category></item><item><title><![CDATA[Dif-CDFusion: A Diffusion-BasedCommon–Differential Network for Infrared andVisible Image Fusion论文分享]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/167703231</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/167703231</guid><author>DTGshaodow</author><pubDate>Mon, 19 Oct 2026 21:21:55 +0800</pubDate><description><![CDATA[原文链接：https://doi.org/10.1109/tgrs.2025.3591809author={Liu, Guanyu and Zhang, Ruiheng and Xu, Lixin and Zhang, Qi and Zhou, Daming}Abstract红外(IR)和可见光(VIS)图像融合旨在通过融合互补传感器数据来增强场景表示。然而，现有的方法不能协调光谱保真度与结构一致性。一方面，灰度融合方法通过丢弃颜色信息来保留结构细节，固有地牺牲了光谱保真度。另一方面，颜色融合技术保持了光谱]]></description><category></category></item><item><title><![CDATA[CFMW: Cross-modality Fusion Mamba for Multispectral Object Detection under Adverse Weather Condition]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/167699993</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/167699993</guid><author>DTGshaodow</author><pubDate>Wed, 14 Oct 2026 20:51:09 +0800</pubDate><description><![CDATA[原文链接：https://doi.org/10.1109/tcsvt.2025.3587918author={Li, Haoyuan and Hu, Qi and Zhou, Binjia and Yao, You and Lin, Jiacheng and Yang, Kailun and Chen, Peng}ABSTRACT融合了可见光-红外光谱线索的跨通道图像可以为目标检测提供更丰富的补充信息。尽管如此，现有的可见光-红外目标探测方法在恶劣天气条件下严重退化。这是因为可见光图像对雨、霾和雪等环境干扰]]></description><category></category></item><item><title><![CDATA[《Mask-DiFuser: A Masked Diffusion Model forUnified Unsupervised Image Fusion》论文分享（侵删）]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/166834836</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/166834836</guid><author>DTGshaodow</author><pubDate>Mon, 12 Oct 2026 11:31:25 +0800</pubDate><description><![CDATA[原文链接：ieeexplore.ieee.org/document/11162636/author={Tang, Linfeng and Li, Chunyu and Ma, Jiayi}Abstract在大多数融合任务中，基本事实(GT)的缺乏给模型的优化、评估和推广带来了巨大的挑战。现有的实现互补上下文聚合的融合方法主要依赖于手工制定的融合规则和复杂的损失函数，这引入了主观性，往往无法适应复杂的现实场景。为了应对这一挑战，我们提出了一种新的融合范例MASK-DiFuser，它巧妙地将无监督图像融合任务转]]></description><category></category></item><item><title><![CDATA[《Multi-modality Image Fusion under Adverse Weather: Mask-Guided Feature Restoration and Interaction》]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/166794448</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/166794448</guid><author>DTGshaodow</author><pubDate>Sat, 10 Oct 2026 18:06:38 +0800</pubDate><description><![CDATA[然而，由于公式4的分母中的亮度偏差是由来自可见图像而不是语义内容的退化因素驱动的，所以退化信息主导了加权，违反了掩模的去耦合原则，并且阻止了它捕捉实际的模式分布。现有的理想条件下的研究主要集中在跨通道信息交互[12，21，28，46]，统一融合框架设计[4，19，37，49]，以及任务驱动优化[1，20，32，35，39]。然而，虽然这种方法降低了优化的复杂性，但它不能完全解决复杂场景下的融合挑战，因为现有方法从干净的图像生成的伪地面真实仍然存在信息损失和通道偏差。公式3定义了传统图像融合的合成范例。]]></description><category></category></item><item><title><![CDATA[《Pushing the Boundaries of Salient Object Detection:A Denoising-Driven Approach》论文分享（侵删）]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/166457204</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/166457204</guid><author>DTGshaodow</author><pubDate>Fri, 09 Oct 2026 20:30:31 +0800</pubDate><description><![CDATA[原文链接：ieeexplore.ieee.org/document/11031107/author={Song, Mengke and Li, Luming and Yu, Xu and Chen, Chenglizhao}摘要显著目标检测旨在识别图像中最引人注目的区域，并将显著目标与其背景区分开来。目前的SOD方法主要使用区分方法，这种方法对于清晰的图像效果很好，但在对象和背景之间具有相似颜色和纹理的复杂场景中很难实现。为了解决这些局限性，我们引入了基于扩散的显著对象检测模型(DiffSOD)，该模型利用]]></description><category></category></item><item><title><![CDATA[《Unaligned UAV RGBT Tracking: A Largescale Benchmark and A Novel Approach》论文分享（侵删）]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/163282296</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/163282296</guid><author>DTGshaodow</author><pubDate>Wed, 23 Sep 2026 09:34:15 +0800</pubDate><description><![CDATA[原文链接：Unaligned UAV RGBT Tracking: A Largescale Benchmark and a Novel Approach| Proceedings of the AAAI Conference on Artificial Intelligenceauthor={Xiao, Yun and Wang, Yuhang and Jin, Jiandong and Zhang, Wankang and Li, Chenglong}摘要随着低空经济的快速发展，无人机场景下的多通道视觉]]></description><category></category></item><item><title><![CDATA[《Multi-modality Image Fusion under Adverse Weather: Mask-Guided Feature Restoration and Interaction》]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/162738870</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/162738870</guid><author>DTGshaodow</author><pubDate>Sat, 25 Jul 2026 17:56:18 +0800</pubDate><description><![CDATA[然而，由于公式4的分母中的亮度偏差是由来自可见图像而不是语义内容的退化因素驱动的，所以退化信息主导了加权，违反了掩模的去耦合原则，并且阻止了它捕捉实际的模式分布。现有的理想条件下的研究主要集中在跨通道信息交互[12，21，28，46]，统一融合框架设计[4，19，37，49]，以及任务驱动优化[1，20，32，35，39]。然而，虽然这种方法降低了优化的复杂性，但它不能完全解决复杂场景下的融合挑战，因为现有方法从干净的图像生成的伪地面真实仍然存在信息损失和通道偏差。公式3定义了传统图像融合的合成范例。]]></description><category></category></item><item><title><![CDATA[《MambaVT: Spatio-Temporal Contextual Modeling for robust RGB-T Tracking》论文分享（侵删）]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/161899202</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/161899202</guid><author>DTGshaodow</author><pubDate>Tue, 23 Jun 2026 11:01:49 +0800</pubDate><description><![CDATA[原文链接：MambaVT: Spatio-Temporal Contextual Modeling for Robust RGB-T Tracking | IEEE Journals & Magazine | IEEE Xplore author={Lai, Simiao and Liu, Chang and Zhu, Jiawen and Kang, Ben and Liu, Yang and Wang, Dong and Lu, Huchuan}Abstract        现有的RGB-T跟踪算法通]]></description><category></category></item><item><title><![CDATA[RSGMamba Reliability-Aware Self-Gated State Space Model for Multimodal Semantic Segmentation论文分享（侵删）]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/161898114</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/161898114</guid><author>DTGshaodow</author><pubDate>Tue, 23 Jun 2026 11:01:34 +0800</pubDate><description><![CDATA[原文链接：[2604.12319] RSGMamba: Reliability-Aware Self-Gated State Space Model for Multimodal Semantic Segmentation author={Xu, Guoan and Xiao, Yang and Gao, Guangwei and Zhu, Dongchen and Qi, Guo-Jun and Jia, Wenjing}摘要多通道语义分割通过利用来自多个传感通道(例如，RGB、深度和温度)的互补信息，已]]></description><category></category></item><item><title><![CDATA[《Deformable Cross-Attention Transformer for Weakly Aligned RGB–T Pedestrian Detection》论文分享（侵删）]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/161259287</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/161259287</guid><author>DTGshaodow</author><pubDate>Wed, 27 May 2026 09:51:34 +0800</pubDate><description><![CDATA[原文链接：https://doi.org/10.1109/tmm.2025.3543056author={Yu Hu and Xiaobo Chen and Sheng Wang and Luyang Liu and Hengyang Shi and Lihong Fan and Jing Tian and Jun Liang}Abstract行人检测在自动驾驶系统中起着至关重要的作用。为了在具有挑战性的条件下确保可靠和有效的检测，研究人员提出了RGB-T(RGB-热)检测器，它将热图像与彩色图像相结合，以]]></description><category></category></item><item><title><![CDATA[《DAMSDet: Dynamic Adaptive Multispectral Detection Transformer...Adaptive Feature Fusion》论文分享(侵删)]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/160419107</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/160419107</guid><author>DTGshaodow</author><pubDate>Fri, 24 Apr 2026 09:46:00 +0800</pubDate><description><![CDATA[原文链接：https://doi.org/10.1007/978-3-031-73383-3_27author={Junjie Guo and Chenqiang Gao and Fangcen liu and Deyu Meng and Xinbo Gao}摘要红外可见光目标检测的目标是通过融合红外和可见光图像的互补信息来实现稳健甚至全天候的目标检测。然而，高度动态变化的互补特征和普遍存在的通道错位使得互补信息的融合变得困难。在本文中，我们提出了一种动态自适应多光谱检测转换器(DAMSDet)来同时解决这]]></description><category></category></item><item><title><![CDATA[Weakly Misalignment-free Adaptive Feature Alignment for UAVs-basedMultimodal Object Detection分享（侵删）]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/160382286</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/160382286</guid><author>DTGshaodow</author><pubDate>Wed, 22 Apr 2026 22:09:19 +0800</pubDate><description><![CDATA[原文链接：https://doi.org/10.1109/cvpr52733.2024.02534author={Chen Chen and Jiahao Qi and Xingyue Liu and Kangcheng Bin and Ruigang Fu and Xikun Hu and Ping Zhong}摘要可见光-红外(RGB-IR)图像融合在无人机目标检测中显示出巨大的潜力。然而，多模式图像对之间的弱不对齐问题限制了它在目标检测中的性能。现有的大多数方法往往忽视通道间隙，而强调严格的比对，导致]]></description><category></category></item><item><title><![CDATA[《SM3Det: A Unified Model for Multi-Modal Remote Sensing Object Detection》论文分享（侵删）]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/160347303</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/160347303</guid><author>DTGshaodow</author><pubDate>Tue, 21 Apr 2026 21:40:25 +0800</pubDate><description><![CDATA[原文链接：https://doi.org/10.1609/aaai.v40i8.37603author={Yuxuan Li and Xiang Li and Yunheng Li and Yicheng Zhang and Yimian Dai and Qibin Hou and Ming-Ming Cheng and Jian Yang}摘要随着遥感技术的快速发展，高分辨率多模式图像的获取越来越广泛。传统的目标检测模型是在单个数据集上训练的，通常限于特定的成像形态和注释格式。然而，这种方法忽略了跨多个通]]></description><category></category></item><item><title><![CDATA[《Improving RGB-infrared object detection with cascade alignment-guided transformer》论文分享（侵删）]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/160223933</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/160223933</guid><author>DTGshaodow</author><pubDate>Mon, 20 Apr 2026 16:56:46 +0800</pubDate><description><![CDATA[原文链接：https://doi.org/10.1016/j.inffus.2024.102246author={Maoxun Yuan and Xiaorong Shi and Nan Wang and Yinyan Wang and Xingxing Wei}摘要多光谱数据在目标检测中的融合，特别是可见光和红外图像，是近年来备受关注的课题。来自可见光(RGB)和红外(IR)图像的互补信息可以改善可变光照条件带来的挑战，使其成为许多领域的宝贵资源，包括RGB-IR目标检测、RGB-IR语义分割和RGB-I]]></description><category></category></item><item><title><![CDATA[《C2Former: Calibrated and Complementary Transformer for RGB-Infrared Object Detection》论文分享（侵删）]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/160142250</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/160142250</guid><author>DTGshaodow</author><pubDate>Thu, 16 Apr 2026 14:55:31 +0800</pubDate><description><![CDATA[原文链接：https://doi.org/10.1109/tgrs.2024.3376819author={Maoxun Yuan and Xingxing Wei}摘要可见光(RGB)和红外(IR)图像上的目标检测作为一种新兴的解决方案，为全天候应用提供了可靠的检测手段，近年来受到了广泛的关注。在红外图像的帮助下，利用RGB-IR组合信息，目标检测器在实际应用中更加可靠和健壮。然而，现有的方法仍然存在着模态误校和融合精度不高的问题。由于Transform对不同特征之间的成对相关性建模能力很强，本文提出了一]]></description><category></category></item><item><title><![CDATA[《IGIANet: Illumination Guided Implicit Alignment Network for Infrared–Visible UAV Detection》论文分享（侵删）]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/159981166</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/159981166</guid><author>DTGshaodow</author><pubDate>Tue, 14 Apr 2026 11:44:04 +0800</pubDate><description><![CDATA[2025)，C2 Former(袁和魏，2024)，SLBAF-Net(程等人2023)、CoDAF(宗珍等人2025)、MBNet(周、陈、曹2020)、慈安(张等2019a)，以及TSFADet(袁，Wang，魏，2022)，我们的方法取得了优越的性能。广泛的实验表明，IGIANet在各种基准测试上的表现优于最先进的模型，在DroneVehicle上获得了80.9%的MAP，在VEDAI上获得了57.1%的MAP，在FLIR上获得了49.4%的MAP。Am(u，v)是对应的频域调制权重映射。]]></description><category></category></item><item><title><![CDATA[《Dark-ISP: Enhancing RAW Image Processing for Low-Light Object Detection》论文分享（侵删）]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/159736608</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/159736608</guid><author>DTGshaodow</author><pubDate>Wed, 08 Apr 2026 18:57:17 +0800</pubDate><description><![CDATA[原文链接：https://arxiv.org/abs/2509.09183author={Jiasheng Guo and Xin Gao and Yuxiang Yan and Guanghao Li and Jian Pu}摘要弱光目标检测对于许多现实世界的应用是至关重要的，但由于图像质量下降，仍然具有挑战性。虽然最近的研究表明，原始图像比RGB图像具有更好的潜力，但现有的方法要么使用具有信息损失的原始RGB图像1，要么使用复杂的框架。为了解决这些问题，我们提出了一个轻量级的自适应图像信号处理(Isp)]]></description><category></category></item><item><title><![CDATA[《Unified Adversarial Patch for Cross-modal Attacks in the Physical World》论文分享（侵删）]]></title><link>https://blog.youkuaiyun.com/DTGshaodow/article/details/159652961</link><guid>https://blog.youkuaiyun.com/DTGshaodow/article/details/159652961</guid><author>DTGshaodow</author><pubDate>Wed, 01 Apr 2026 21:45:55 +0800</pubDate><description><![CDATA[原文链接：https://doi.org/10.1109/tpami.2023.3330769author={Xingxing Wei and Yao Huang and Yitong Sun and Jie Yu}摘要        最近，为了躲避基于DNN的对象检测器，已经出现了物理对抗性攻击。为了确保安全，多个场景同时部署了可见光传感器和红外传感器，导致这些单模物理攻击失败。为了显示这种场景下的潜在风险，我们提出了一个统一的对抗性补丁来执行跨模式的物理攻击，即通过一个补丁同时欺骗可见光和红外目标探测器]]></description><category></category></item></channel></rss>