墙裂推荐：想获取更多前沿论文及算法优化idea冲击顶会或发表专利，包含目标检测、目标跟踪、图像分割、视频分割、Visual Grounding、可见光红外融合、多任务学习、多模态基础模型、文生图、自动驾驶、BEV、占用预测、具身智能VLA、深度估计、动作识别、表情识别、三维重建、点云3D检测、医学图像分割、医学图像目标检测、医学大模型、缺陷检测、异常检测、遥感图像分割、遥感图像变化检测、数字人、知识蒸馏、视频理解、3D生成、姿态估计、图像增强、人群/目标计数、视频编辑、图像去雨等众多主题，请参考：https://qcno08je5sgu.feishu.cn/

1.【图像融合】UniFusion: A Unified Image Fusion Framework with Robust Representation and Source-Aware Preservation

• 论文地址：https://arxiv.org//pdf/2603.14214
• 开源代码：https://github.com/dusongcheng/UniFusion

2.【多模态大模型】UAVBench and UAVIT-1M: Benchmarking and Enhancing MLLMs for Low-Altitude UAV Vision-Language Understanding

• 论文地址：https://arxiv.org//pdf/2603.14336
• 工程主页：SOCIAL MEDIA TITLE TAG
• 开源代码：https://github.com/ZhanYang-nwpu/UAVBench-and-UAVIT-1M

3.【多模态大模型】Deeper Thought, Weaker Aim: Understanding and Mitigating Perceptual Impairment during Reasoning in Multimodal Large Language Models

• 论文地址：https://arxiv.org//pdf/2603.14184
• 开源代码（即将开源）：https://github.com/Ivine11/VRGA

4.【医学大模型】（ICLR2026）How Do Medical MLLMs Fail? A Study on Visual Grounding in Medical Images

• 论文地址：https://arxiv.org//pdf/2603.14323
• 工程主页：How Do Medical MLLMs Fail? A Study on Visual Grounding in Medical Images
• 开源代码：https://github.com/Guimeng-Leo-Liu/Medical-MLLMs-Fail

5.【行人重识别】（CVPR2026）BIT: Matching-based Bi-directional Interaction Transformation Network for Visible-Infrared Person Re-Identification

• 论文地址：https://arxiv.org//pdf/2603.14243
• 开源代码（即将开源）：https://github.com/Xuan266/BIT

6.【数字人】AvatarForcing: One-Step Streaming Talking Avatars via Local-Future Sliding-Window Denoising

• 论文地址：https://arxiv.org//pdf/2603.14331
• 工程主页：https://cuiliyuan121.github.io/AvatarForcing/
• 开源代码：https://github.com/KlingAIResearch/AvatarForcing/tree/main

7.【视觉语言导航】AerialVLA: A Vision-Language-Action Model for UAV Navigation via Minimalist End-to-End Control

• 论文地址：https://arxiv.org//pdf/2603.14363
• 开源代码：https://github.com/XuPeng23/AerialVLA

8.【视觉语言导航】（ICLR2026）All-day Multi-scenes Lifelong Vision-and-Language Navigation with Tucker Adaptation

• 论文地址：https://arxiv.org//pdf/2603.14276
• 工程主页：All-Day Multi-Scenes Lifelong Vision-And-Language Navigation With Tucker-Adaption
• 开源代码：https://github.com/Ganvin-Li/AlldayWalker

9.【文生图】Fair Benchmarking of Emerging One-Step Generative Models Against Multistep Diffusion and Flow Models

• 论文地址：https://arxiv.org//pdf/2603.14186
• 开源代码：https://github.com/Harvard-AI-and-Robotics-Lab/FairBenchmarkingFlow

10.【文生视频】Early Failure Detection and Intervention in Video Diffusion Models

• 论文地址：https://arxiv.org//pdf/2603.14320
• 开源代码（即将开源）：https://github.com/kaist-ami/Early-failure-video-diffusion

11.【文生视频】Seeking Physics in Diffusion Noise

• 论文地址：https://arxiv.org//pdf/2603.14294
• 工程主页：Seeking Physics in Diffusion Noise
• 代码即将开源

12.【图像生成】Representation Alignment for Just Image Transformers is not Easier than You Think

• 论文地址：https://arxiv.org//pdf/2603.14366
• 开源代码：https://github.com/kaist-cvml/PixelREPA

群内包含目标检测、图像分割、目标跟踪、Transformer、多模态、NeRF、GAN、缺陷检测、显著目标检测、关键点检测、超分辨率重建、SLAM、人脸、OCR、生物医学图像、三维重建、姿态估计、自动驾驶感知、深度估计、视频理解、行为识别、图像去雾、图像去雨、图像修复、图像检索、车道线检测、点云目标检测、点云分割、图像压缩、运动预测、神经网络量化、网络部署等多个领域的大佬，不定期分享技术知识、面试技巧和内推招聘信息。