即插即用型 - 分类 - Zhaoylee's Blogs https://zhaoylee.github.io/Blogs_lovelt/categories/%E5%8D%B3%E6%8F%92%E5%8D%B3%E7%94%A8%E5%9E%8B/ 即插即用型 - 分类 - Zhaoylee's Blogs Hugo -- gohugo.iozh-CNSat, 04 Apr 2026 12:31:36 +0800 MONOLSS Learnable Sample Selection For Monocular 3D Detection https://zhaoylee.github.io/Blogs_lovelt/posts/monolss--learnable-sample-selection-for-monocular-3d-detection/ Sat, 04 Apr 2026 12:31:36 +0800 zhaoylee https://zhaoylee.github.io/Blogs_lovelt/posts/monolss--learnable-sample-selection-for-monocular-3d-detection/ 博客的简述]]> Mix-Teaching: A Simple, Unified and Effective Semi-Supervised Learning Framework for Monocular 3D Object Detection https://zhaoylee.github.io/Blogs_lovelt/posts/mix-teaching--a-simple-unified-and-effective--semi-supervised-learning-framework-for--monocular-3d-object-detection/ Mon, 30 Mar 2026 23:17:02 +0800 zhaoylee https://zhaoylee.github.io/Blogs_lovelt/posts/mix-teaching--a-simple-unified-and-effective--semi-supervised-learning-framework-for--monocular-3d-object-detection/ 本文提出 Mix-Teaching，首个专为单目 3D 目标检测设计的半监督学习统一框架。针对伪标签“低精度与低召回率”导致的确认偏差痛点，创新性提出“分解-重组”的跨帧实例级混合机制，并结合基于不确定性的过滤策略，优雅且高效地释放了无标注数据的潜力。]]> Adaptive Dual Uncertainty Optimization: Boosting Monocular 3D Object Detection under Test Time Shifts https://zhaoylee.github.io/Blogs_lovelt/posts/adaptive-dual-uncertainty-optimization---boosting-monocular-3d-object-detection-under-test-time-shifts/ Mon, 30 Mar 2026 10:44:00 +0800 zhaoylee https://zhaoylee.github.io/Blogs_lovelt/posts/adaptive-dual-uncertainty-optimization---boosting-monocular-3d-object-detection-under-test-time-shifts/ 针对 M3OD 在未知域“测试时偏移”导致的性能断崖，本文提出双重不确定性优化 (DUO)。核心是通过无监督 Focal Loss 压制语义模糊，并用语义感知法向量约束修复空间几何坍塌，大幅提升落地鲁棒性。]]> OBMO: One Bounding Box Multiple Objects for Monocular 3D Object Detection https://zhaoylee.github.io/Blogs_lovelt/posts/obmo-one-bounding-box-multiple-objects-for-monocular-3d-object-detection/ Tue, 24 Mar 2026 10:48:33 +0800 zhaoylee https://zhaoylee.github.io/Blogs_lovelt/posts/obmo-one-bounding-box-multiple-objects-for-monocular-3d-object-detection/ 简述:一个框对应多个物体位置，通过施加软标签，帮助网络训练稳定，从而提升一定的性能;]]> OCM3D: Object-Centric Monocular 3D Object Detection https://zhaoylee.github.io/Blogs_lovelt/posts/ocm3d--object-centric-monocular-3d-object-detection/ Mon, 16 Mar 2026 09:12:18 +0800 zhaoylee https://zhaoylee.github.io/Blogs_lovelt/posts/ocm3d--object-centric-monocular-3d-object-detection/

🏛️ 会议/期刊：arxiv
📅 发表年份：2021
💻 开源代码：OBMO_GUPNet
📄 论文题目：OCM3D: Object-Centric Monocular 3D Object Detection

1. 文献背景、研究目的与核心问题

  • 研究背景：单目 3D 目标检测（Monocular 3D Object Detection）是一个高度病态（ill-posed）的问题。主流方法通常依赖纯图像或将其转化为伪激光雷达（Pseudo-LiDAR）点云。然而，前者难以捕捉像素间的 3D 空间几何关系，后者则受困于单目深度估计带来的巨大点云噪声。

    ]]> LR3D: Improving Distant 3D Object Detection Using 2D Box Supervision https://zhaoylee.github.io/Blogs_lovelt/posts/lr3d--improving-distant-3d-object-detection-using-2d-box-supervision/ Sun, 15 Mar 2026 22:23:00 +0800 zhaoylee https://zhaoylee.github.io/Blogs_lovelt/posts/lr3d--improving-distant-3d-object-detection-using-2d-box-supervision/

    🏛️ 会议/期刊：CVPR
    📅 发表年份：2024
    💻 开源代码：无
    📄 论文题目：Improving Distant 3D Object Detection Using 2D Box Supervision

    这篇由 NVIDIA 等机构的研究人员发表在 CVPR 2024 的重磅论文 《Improving Distant 3D Object Detection Using 2D Box Supervision》(简称 LR3D)，切入了一个目前高阶自动驾驶极其头疼的落地难题：远距离感知（Long-Range Detection）。它展示了如何用最廉价的标注，榨取单目视觉在远距离上的极限潜力。

    ]]>         OBMO: One Bounding Box Multiple Objects for Monocular 3D Object Detection https://zhaoylee.github.io/Blogs_lovelt/posts/obmo--one-bounding-box-multiple-objects-for-monocular-3d-object-detection/ Sun, 15 Mar 2026 21:59:12 +0800 zhaoylee https://zhaoylee.github.io/Blogs_lovelt/posts/obmo--one-bounding-box-multiple-objects-for-monocular-3d-object-detection/

    🏛️ 会议/期刊：IEEE TIP
    📅 发表年份：2023
    💻 开源代码：mrsempress/OBMO_patchnet
    📄 论文题目：OBMO: One Bounding Box Multiple Objects for Monocular 3D Object Detection

    这篇发表于 IEEE TIP (2023) 的经典论文 《OBMO: One Bounding Box Multiple Objects for Monocular 3D Object Detection》 切入点非常犀利。它没有在复杂的网络主干上做文章，而是直击单目 3D 目标检测在“底层数学物理逻辑”上的痛点，提出了一种极其优雅的“即插即用（Plug-and-play）”训练策略。

    ]]>         MonoXiver： Monocular 3D Object Detection with Bounding Box Denoising in 3D by Perceiver https://zhaoylee.github.io/Blogs_lovelt/posts/monoxiver--monocular-3d-object-detection-with-bounding-box-denoising-in-3d-by-perceivermonocular-3d-object-detection-with-bounding-box-denoising-in-3d-by-perceiver/ Sun, 15 Mar 2026 21:14:36 +0800 zhaoylee https://zhaoylee.github.io/Blogs_lovelt/posts/monoxiver--monocular-3d-object-detection-with-bounding-box-denoising-in-3d-by-perceivermonocular-3d-object-detection-with-bounding-box-denoising-in-3d-by-perceiver/

    🏛️ 会议/期刊：ICCV
    📅 发表年份：2023
    💻 开源代码：Xianpeng919/monoxiver (ICCV'23)
    📄 论文题目：Monocular 3D Object Detection with Bounding Box Denoising in 3D by Perceiver

    这篇发表于 ICCV 2023 的论文 《Monocular 3D Object Detection with Bounding Box Denoising in 3D by Perceiver》(简称 MonoXiver)，提供了一个非常经典且极具工程价值的“自上而下（Top-down）”纠错思路。它并没有试图发明一种全新的主干网络，而是设计了一个强大的“插件”，专门用来拯救那些定位不准的预测框。

    ]]>         PLOT: Pseudo-Labeling via Object Tracking for Monocular 3D Object Detection https://zhaoylee.github.io/Blogs_lovelt/posts/plot---pseudo-labeling-via-object-tracking-for-monocular-3d-object-detection/ Sun, 15 Mar 2026 20:52:51 +0800 zhaoylee https://zhaoylee.github.io/Blogs_lovelt/posts/plot---pseudo-labeling-via-object-tracking-for-monocular-3d-object-detection/

    🏛️ 会议/期刊：ICLR
    📅 发表年份：2026
    💻 开源代码：无
    📄 论文题目：PLOT: Pseudo-Labeling via Object Tracking for Monocular 3D Object Detection

    一、 背景、研究目的与核心问题

    • 研究背景： 单目 3D 目标检测模型极度“吃数据”。然而，人工标注 3D 边界框极其昂贵且耗时，导致目前带 3D 标签的数据集规模很小，严重限制了模型的泛化能力。

      ]]> Object-Scene-Camera Decomposition and Recomposition for Data-Efficient Monocular 3D Object Detection https://zhaoylee.github.io/Blogs_lovelt/posts/dr-traversal-m3d--object-scene-camera-decomposition-and-recomposition-for-data-efficient-monocular-3d-object-detection/ Sun, 15 Mar 2026 19:59:15 +0800 zhaoylee https://zhaoylee.github.io/Blogs_lovelt/posts/dr-traversal-m3d--object-scene-camera-decomposition-and-recomposition-for-data-efficient-monocular-3d-object-detection/

      🏛️ 会议/期刊：IJCV
      📅 发表年份：2026
      💻 开源代码：GitHub 链接
      📄 论文题目：Object-Scene-Camera Decomposition and Recomposition for Data-Efficient Monocular 3D Object Detection

      一、 背景、研究目的与核心问题

      • 研究背景： 单目 3D 目标检测本身是一个病态（ill-posed）问题，由于缺乏深度信息，模型极度依赖海量、多样化的训练数据来学习鲁棒的特征。

        ]]>