Skip to content
View lartpang's full-sized avatar
😶‍🌫️
Completing a Ph.D...
😶‍🌫️
Completing a Ph.D...

Block or report lartpang

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Please don't include any personal information such as legal names or email addresses. Maximum 100 characters, markdown supported. This note will be visible to only you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
lartpang/README.md

Hi 👋, I'm lartpang

🧑‍🤝‍🧑 Me

$$ \textbf{life} = \int_{birth}^{now} \mathbf{happy}(time) + \mathbf{sad}(time) d(time) $$

A Python and PyTorch developer, deep-learning worker and open-source activist.

Created by the awesome tool. 😊

📝 Recent Writing

  • 生成模型 | 扩散模型损失函数公式推导 - Sat, 23 Aug 2025: 本文推导了扩散模型的损失函数,通过引入前向分布简化计算,最终将损失分解为三部分:$L_T$(可忽略的常量)、$L_{t-1}$(KL散度项)和$L_0$(重构误差)。
  • 生成模型 | 扩散模型公式推导 - Sat, 23 Aug 2025: 本文介绍了扩散模型的前向加噪和反向去噪过程。前向过程通过马尔科夫链逐步将数据$x_0$转化为高斯噪声$x_T$,其中噪声强度由预设参数$�eta_t$控制。反向过程则利用神经网络从噪声$x_T$逐步恢复原始数据$x_0$。
  • ICCV 2025 | Reverse Convolution and Its Applications to Image Restoration - Sun, 17 Aug 2025: 本文提出了一种新颖的深度可分离反向卷积算子(reverse convolution),通过建立并求解正则化最小二乘优化问题,实现了对depthwise卷积的有效反转。该算子采用FFT推导闭式解,并详细研究了核初始化、padding策略等实现细节。基于此构建的reverse卷积块结合了层归一化、1×1卷积和GELU激活,形成类Transformer结构,可直接替换现有网络中的常规卷积层,构建ConverseNet。
  • TCSVT 2023 | StructToken - Rethinking Semantic Segmentation with Structural Prior - Sun, 17 Aug 2025: 一种新的语义分割范式,通过结构化token直接构建语义掩码并逐步细化,而非传统逐像素分类方法。作者设计了三种交互结构(CSE、SSE和静态卷积)来捕获特征图中的结构信息,并通过堆叠处理单元实现mask细化。
  • torchvision 中 deform_conv2d 操作的经验性解析 - Sun, 17 Aug 2025: 详细解析了torchvision中可变形卷积(deform_conv2d)的实现原理和使用方法。
  • 一次由默认参数引起的思考 - Sun, 17 Aug 2025: 本文探讨了依赖版本更新导致代码输出不一致的问题。作者在迁移代码时发现,由于Pillow图像处理库从6.2.1升级到7.2.0,其默认插值策略改变导致resize()函数输出结果不同。文章分析了默认参数的利弊,指出其虽提升开发效率但存在潜在风险。作者建议采取两种应对策略:一是固定依赖版本确保稳定性;二是对关键参数进行显式配置。最后强调开发应以程序稳定运行为首要目标,盲目追求新版本可能得不偿失,并提醒开发者需谨慎对待工具依赖的版本管理。
  • TIP 2004 | Image quality assessment: From error visibility to structural similarity - Sun, 17 Aug 2025: 本文介绍了全参考图像质量评估方法SSIM(结构相似性指数)的设计背景与实现。传统评估方法如MSE和PSNR虽计算简单,但与人类感知质量匹配度低。SSIM基于结构信息退化假设,通过亮度、对比度和结构三个分量评估图像质量。论文详细阐述了SSIM的算法框架,并对比了不同实现的高斯滤波处理方式差异。作者基于PyTorch实现了可微分的MSSIM代码,支持用户自定义padding和核形式参数,确保与现有实现兼容。该指标在图像处理系统优化、算法评估等领域具有重要应用价值。
  • ACMMM 2024 | Wave-Mamba: Wavelet State Space Model for Ultra-High-Definition Low-Light Image Enhance - Fri, 01 Aug 2025: 针对超高清低照度图像增强中的计算复杂度和信息丢失问题,提出Wave-Mamba模型。该模型创新性地结合离散小波变换(DWT)与状态空间模型(SSM),通过小波域分析发现:1)93.7%图像能量集中于低频分量;2)高频对增强结果影响微弱。基于此,设计低频状态空间模块(LFSSBlock)进行全局增强,并通过改进的高频增强模块(HFEBlock)校正细节。
  • ICCV 2025 | WaveMamba: Wavelet-Driven Mamba Fusion for RGB-Infrared Object Detection - Fri, 01 Aug 2025: 本文提出WaveMamba,一种基于小波变换和Mamba的RGB-红外跨模态目标检测方法。研究发现RGB和红外图像在频域具有互补特性:红外图像低频信息丰富,RGB图像高频细节突出。WaveMamba通过离散小波变换分解特征,采用低频Mamba融合块(结合通道交换和门控注意力)和高频绝对最大值增强策略,实现高效特征融合。在六个基准数据集上的实验表明,该方法平均mAP提升4.5%,同时保持较低计算开销,为跨模态目标检测提供了新思路。
  • ICCV 2025 | CWNet: Causal Wavelet Network for Low-Light Image Enhancement - Thu, 24 Jul 2025: 本文提出一种基于因果推理与小波变换的低光照图像增强方法。CWNet通过因果干预分析揭示潜在因果关系,采用全局度量学习分离因果/非因果因子,并引入实例级CLIP语义损失确保局部一致性。同时设计基于小波变换的主干网络优化频域信息恢复。实验表明,CWNet在多个数据集上优于现有方法,有效解决了光照不均与语义保持的挑战。该方法为低光增强提供了新的因果推理视角,显著提升了视觉质量与语义准确性。

View the archives @ csdn@p_lart.

📽️ Some Projects

Name Stars Description
Hands-on-Docker (中文) stars 一份详尽的 Docker 使用指南。
Awesome-Class-Activation-Map stars An awesome list of papers and tools about the class activation map (CAM) technology.
PyTorchTricks stars Some tricks of pytorch…
MethodsCmp stars A Simple Toolkit for Counting the FLOPs/MACs, Parameters and FPS of Pytorch-based Methods.
PySODEvalToolkit stars A Python-based salient object detection and video object segmentation evaluation toolbox.
PySODMetrics stars A simple and efficient implementation of SOD metrcis.
PyLoss stars Some loss functions for deeplearning.
OpticalFlowBasedVOS stars A simple and efficient codebase for the optical flow based video object segmentation.
CoSaliencyProj stars A project for co-saliency detection. Some codes are borrowed from ICNet. Thanks to ICNet Intra-saliency Correlation Network for Co-Saliency Detection (NIPS2020)
RunIt stars A simple program scheduler for your code on different devices.
RegisterIt stars Register it: A more flexible register for the DeepLearning project.
mssim.pytorch stars A better pytorch-based implementation for the mean structural similarity. Differentiable simpler SSIM and MS-SSIM.
tta.pytorch stars Test-Time Augmentation library for Pytorch.
YuQueTools stars A simple tool to download your own articles from yuque.
ManageMyAttachments stars Manage the attachments of your own obsidian vault.

Pinned Loading

  1. awesome-segmentation-saliency-dataset awesome-segmentation-saliency-dataset Public

    A collection of some datasets for segmentation / saliency detection. Welcome to PR...:smile:

    586 96

  2. PyTorchTricks PyTorchTricks Public

    Some tricks of pytorch... ⭐

    1.2k 128

  3. SAMs-CDConcepts-Eval SAMs-CDConcepts-Eval Public

    Inspiring the Next Generation of Segment Anything Models: Comprehensively Evaluate SAM and SAM 2 with Diverse Prompts Towards Context-Dependent Concepts under Different Scenes

    Python 399 12

  4. ZoomNeXt ZoomNeXt Public

    ZoomNeXt: A Unified Collaborative Pyramid Network for Camouflaged Object Detection (TPAMI 2024)

    Python 55 6

  5. PySODMetrics PySODMetrics Public

    PySODMetrics: A Simple and Efficient Implementation of Grayscale/Binary Segmentation Metrcis

    Python 181 21

  6. OVCamo OVCamo Public

    (ECCV 2024) Open-Vocabulary Camouflaged Object Segmentation

    Python 27 1