[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-shayneobrien--generative-models":3,"tool-shayneobrien--generative-models":61},[4,18,26,36,44,53],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":17},4358,"openclaw","openclaw\u002Fopenclaw","OpenClaw 是一款专为个人打造的本地化 AI 助手，旨在让你在自己的设备上拥有完全可控的智能伙伴。它打破了传统 AI 助手局限于特定网页或应用的束缚，能够直接接入你日常使用的各类通讯渠道，包括微信、WhatsApp、Telegram、Discord、iMessage 等数十种平台。无论你在哪个聊天软件中发送消息，OpenClaw 都能即时响应，甚至支持在 macOS、iOS 和 Android 设备上进行语音交互，并提供实时的画布渲染功能供你操控。\n\n这款工具主要解决了用户对数据隐私、响应速度以及“始终在线”体验的需求。通过将 AI 部署在本地，用户无需依赖云端服务即可享受快速、私密的智能辅助，真正实现了“你的数据，你做主”。其独特的技术亮点在于强大的网关架构，将控制平面与核心助手分离，确保跨平台通信的流畅性与扩展性。\n\nOpenClaw 非常适合希望构建个性化工作流的技术爱好者、开发者，以及注重隐私保护且不愿被单一生态绑定的普通用户。只要具备基础的终端操作能力（支持 macOS、Linux 及 Windows WSL2），即可通过简单的命令行引导完成部署。如果你渴望拥有一个懂你",349277,3,"2026-04-06T06:32:30",[13,14,15,16],"Agent","开发框架","图像","数据工具","ready",{"id":19,"name":20,"github_repo":21,"description_zh":22,"stars":23,"difficulty_score":10,"last_commit_at":24,"category_tags":25,"status":17},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,"2026-04-05T11:01:52",[14,15,13],{"id":27,"name":28,"github_repo":29,"description_zh":30,"stars":31,"difficulty_score":32,"last_commit_at":33,"category_tags":34,"status":17},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",160784,2,"2026-04-19T11:32:54",[14,13,35],"语言模型",{"id":37,"name":38,"github_repo":39,"description_zh":40,"stars":41,"difficulty_score":32,"last_commit_at":42,"category_tags":43,"status":17},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",109154,"2026-04-18T11:18:24",[14,15,13],{"id":45,"name":46,"github_repo":47,"description_zh":48,"stars":49,"difficulty_score":32,"last_commit_at":50,"category_tags":51,"status":17},6121,"gemini-cli","google-gemini\u002Fgemini-cli","gemini-cli 是一款由谷歌推出的开源 AI 命令行工具，它将强大的 Gemini 大模型能力直接集成到用户的终端环境中。对于习惯在命令行工作的开发者而言，它提供了一条从输入提示词到获取模型响应的最短路径，无需切换窗口即可享受智能辅助。\n\n这款工具主要解决了开发过程中频繁上下文切换的痛点，让用户能在熟悉的终端界面内直接完成代码理解、生成、调试以及自动化运维任务。无论是查询大型代码库、根据草图生成应用，还是执行复杂的 Git 操作，gemini-cli 都能通过自然语言指令高效处理。\n\n它特别适合广大软件工程师、DevOps 人员及技术研究人员使用。其核心亮点包括支持高达 100 万 token 的超长上下文窗口，具备出色的逻辑推理能力；内置 Google 搜索、文件操作及 Shell 命令执行等实用工具；更独特的是，它支持 MCP（模型上下文协议），允许用户灵活扩展自定义集成，连接如图像生成等外部能力。此外，个人谷歌账号即可享受免费的额度支持，且项目基于 Apache 2.0 协议完全开源，是提升终端工作效率的理想助手。",100752,"2026-04-10T01:20:03",[52,13,15,14],"插件",{"id":54,"name":55,"github_repo":56,"description_zh":57,"stars":58,"difficulty_score":32,"last_commit_at":59,"category_tags":60,"status":17},4721,"markitdown","microsoft\u002Fmarkitdown","MarkItDown 是一款由微软 AutoGen 团队打造的轻量级 Python 工具，专为将各类文件高效转换为 Markdown 格式而设计。它支持 PDF、Word、Excel、PPT、图片（含 OCR）、音频（含语音转录）、HTML 乃至 YouTube 链接等多种格式的解析，能够精准提取文档中的标题、列表、表格和链接等关键结构信息。\n\n在人工智能应用日益普及的今天，大语言模型（LLM）虽擅长处理文本，却难以直接读取复杂的二进制办公文档。MarkItDown 恰好解决了这一痛点，它将非结构化或半结构化的文件转化为模型“原生理解”且 Token 效率极高的 Markdown 格式，成为连接本地文件与 AI 分析 pipeline 的理想桥梁。此外，它还提供了 MCP（模型上下文协议）服务器，可无缝集成到 Claude Desktop 等 LLM 应用中。\n\n这款工具特别适合开发者、数据科学家及 AI 研究人员使用，尤其是那些需要构建文档检索增强生成（RAG）系统、进行批量文本分析或希望让 AI 助手直接“阅读”本地文件的用户。虽然生成的内容也具备一定可读性，但其核心优势在于为机器",93400,"2026-04-06T19:52:38",[52,14],{"id":62,"github_repo":63,"name":64,"description_en":65,"description_zh":66,"ai_summary_zh":66,"readme_en":67,"readme_zh":68,"quickstart_zh":69,"use_case_zh":70,"hero_image_url":71,"owner_login":72,"owner_name":73,"owner_avatar_url":74,"owner_bio":75,"owner_company":76,"owner_location":77,"owner_email":78,"owner_twitter":78,"owner_website":79,"owner_url":80,"languages":81,"stars":90,"forks":91,"last_commit_at":92,"license":76,"difficulty_score":32,"env_os":93,"env_gpu":94,"env_ram":93,"env_deps":95,"category_tags":101,"github_topics":102,"view_count":32,"oss_zip_url":78,"oss_zip_packed_at":78,"status":17,"created_at":122,"updated_at":123,"faqs":124,"releases":125},9679,"shayneobrien\u002Fgenerative-models","generative-models","Annotated, understandable, and visually interpretable PyTorch implementations of: VAE, BIRVAE, NSGAN, MMGAN, WGAN, WGANGP, LSGAN, DRAGAN, BEGAN, RaGAN, InfoGAN, fGAN, FisherGAN","generative-models 是一个基于 PyTorch 的开源项目，提供了多种主流生成模型（包括各类 GAN 变体和 VAE）的代码实现。它旨在解决深度学习领域中生成模型种类繁多、理论复杂且代码难以复现的痛点，通过提供带有详细注释和可视化功能的参考代码，让抽象的算法变得直观易懂。\n\n该项目非常适合 AI 研究人员、深度学习开发者以及希望深入理解生成对抗网络原理的学生使用。其核心亮点在于极高的可读性与模块化设计：不同模型间的差异主要体现为损失函数的计算方式，用户只需修改少量代码即可从一种模型（如 NSGAN）快速切换到另一种（如 LSGAN），极大地降低了尝试新算法的门槛。此外，项目内置了丰富的可视化工具，支持潜在空间表示分析和训练过程监控，并自动适配 CPU 或 GPU 环境。无论是用于教学演示、算法对比研究，还是作为开发自定义生成模型的起点，generative-models 都是一个实用且友好的技术资源库。","# Overview\nPyTorch 0.4.1 | Python 3.6.5\n\nAnnotated implementations with comparative introductions for minimax, non-saturating, wasserstein, wasserstein gradient penalty, least squares, deep regret analytic, bounded equilibrium, relativistic, f-divergence, Fisher, and information generative adversarial networks (GANs), and standard, variational, and bounded information rate variational autoencoders (VAEs).\n\nPaper links are supplied at the beginning of each file with a short summary of the paper. See src folder for files to run via terminal, or notebooks folder for Jupyter notebook visualizations via your local browser. The main file changes can be see in the [```train```](https:\u002F\u002Fgithub.com\u002Fshayneobrien\u002Fgenerative-models\u002Fblob\u002Fmaster\u002Fsrc\u002Fns_gan.py#L94-L170), [```train_D```](https:\u002F\u002Fgithub.com\u002Fshayneobrien\u002Fgenerative-models\u002Fblob\u002Fmaster\u002Fsrc\u002Fns_gan.py#L172-L194), and [```train_G```](https:\u002F\u002Fgithub.com\u002Fshayneobrien\u002Fgenerative-models\u002Fblob\u002Fmaster\u002Fsrc\u002Fns_gan.py#L196-L216) of the Trainer class, although changes are not completely limited to only these two areas (e.g. Wasserstein GAN clamps weight in the train function, BEGAN gives multiple outputs from train_D, fGAN has a slight modification in viz_loss function to indicate method used in title).\n\nAll code in this repository operates in a generative, unsupervised manner on binary (black and white) MNIST. The architectures are compatible with a variety of datatypes (1D, 2D, square 3D images). Plotting functions work with binary\u002FRGB images. If a GPU is detected, the models use it. Otherwise, they default to CPU. VAE Trainer classes contain methods to visualize latent space representations (see [```make_all```](https:\u002F\u002Fgithub.com\u002Fshayneobrien\u002Fgenerative-models\u002Fblob\u002Fmaster\u002Fsrc\u002Fvae.py#L333-L343) function).\n\n# Usage\nTo initialize an environment:\n```\npython -m venv env  \n. env\u002Fbin\u002Factivate  \npip install -r requirements.txt  \n```\n\nFor playing around in Jupyer notebooks:\n```\njupyter notebook\n```\n\nTo run from Terminal:\n```\ncd src\npython bir_vae.py\n```\n\n# New Models\n\nOne of the primary purposes of this repository is to make implementing deep generative model (i.e., GAN\u002FVAE) variants as easy as possible. This is possible because, typically but not always (e.g. BIRVAE), the proposed modifications only apply to the way loss is computed for backpropagation. Thus, the core training class is structured in such a way that most new implementations should only require edits to the ```train_D``` and ```train_G``` functions of GAN Trainer classes, and the ```compute_batch``` function of VAE Trainer classes.\n\nSuppose we have a non-saturating GAN and we wanted to implement a least-squares GAN. To do this, all we have to do is change two lines:\n\n[Original](https:\u002F\u002Fgithub.com\u002Fshayneobrien\u002Fgenerative-models\u002Fblob\u002Fmaster\u002Fsrc\u002Fns_gan.py#L166-L208) (NSGAN)\n```\ndef train_D(self, images):\n  ...\n  D_loss = -torch.mean(torch.log(DX_score + 1e-8) + torch.log(1 - DG_score + 1e-8))\n\n  return D_loss\n```\n```\ndef train_G(self, images):\n  ...\n  G_loss = -torch.mean(torch.log(DG_score + 1e-8))\n\n  return G_loss\n```\n\n[New](https:\u002F\u002Fgithub.com\u002Fshayneobrien\u002Fgenerative-models\u002Fblob\u002Fmaster\u002Fsrc\u002Fls_gan.py#L166-L209) (LSGAN)\n```\ndef train_D(self, images):\n  ...\n  D_loss = (0.50 * torch.mean((DX_score - 1.)**2)) + (0.50 * torch.mean((DG_score - 0.)**2))\n\n  return D_loss\n```\n```\ndef train_G(self, images):\n  ...\n  G_loss = 0.50 * torch.mean((DG_score - 1.)**2)\n\n  return G_loss\n```\n\n# Model Architecture\nThe architecture chosen in these implementations for both the generator (G) and discriminator (D) consists of a simple, two-layer feedforward network. While this will give sensible output for MNIST, in practice it is recommended to use deep convolutional architectures (i.e. DCGANs) to get nicer outputs. This can be done by editing the Generator and Discriminator classes for GANs, or the Encoder and Decoder classes for VAEs.\n\n# Visualization\nAll models were trained for 25 epochs with hidden dimension 400, latent dimension 20. Other implementation specifics are as close to the respective original paper (linked) as possible.\n\n*Model* | *Epoch 1* | *Epoch 25* | *Progress* | *Loss*\n:---: | :---: | :---: | :---: | :---: |\n[MMGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1406.2661) | \u003Cimg src = 'viz\u002FMMGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FMMGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FMMGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FMMGAN_loss.png' height = '150px'>\n[NSGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1406.2661) | \u003Cimg src = 'viz\u002FNSGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FNSGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FNSGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FNSGAN_loss.png' height = '150px'>\n[WGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1701.07875) | \u003Cimg src = 'viz\u002FWGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FWGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FWGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FWGAN_loss.png' height = '150px'>\n[WGPGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1704.00028) | \u003Cimg src = 'viz\u002FWGPGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FWGPGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FWGPGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FWGPGAN_loss.png' height = '150px'>\n[DRAGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1705.07215) | \u003Cimg src = 'viz\u002FDRAGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FDRAGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FDRAGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FDRAGAN_loss.png' height = '150px'>\n[BEGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1703.10717) | \u003Cimg src = 'viz\u002FBEGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FBEGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FBEGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FBEGAN_loss.png' height = '150px'>\n[LSGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1611.04076) | \u003Cimg src = 'viz\u002FLSGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FLSGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FLSGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FLSGAN_loss.png' height = '150px'>\n[RaNSGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1807.00734) | \u003Cimg src = 'viz\u002FRaNSGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FRaNSGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FRaNSGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FRaNSGAN_loss.png' height = '150px'>\n[FisherGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1606.07536) | \u003Cimg src = 'viz\u002FFisherGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FFisherGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FFisherGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FFisherGAN_loss.png' height = '150px'>\n[InfoGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1606.03657) | \u003Cimg src = 'viz\u002FInfoGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FInfoGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FInfoGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FInfoGAN_loss.png' height = '150px'>\n[f-TVGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1606.00709) | \u003Cimg src = 'viz\u002FfGAN\u002Ftotal_variation\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FfGAN\u002Ftotal_variation\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FfGAN_total_variation_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FfGAN_total_variation_loss.png' height = '150px'>\n[f-PearsonGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1606.00709) | \u003Cimg src = 'viz\u002FfGAN\u002Fpearson\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FfGAN\u002Fpearson\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FfGAN_pearson_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FfGAN_pearson_loss.png' height = '150px'>\n[f-JSGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1606.00709) | \u003Cimg src = 'viz\u002FfGAN\u002Fjensen_shannon\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FfGAN\u002Fjensen_shannon\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FfGAN_jensen_shannon_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FfGAN_jensen_shannon_loss.png' height = '150px'>\n[f-ForwGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1606.00709) | \u003Cimg src = 'viz\u002FfGAN\u002Fforward_kl\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FfGAN\u002Fforward_kl\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FfGAN_forward_kl_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FfGAN_forward_kl_loss.png' height = '150px'>\n[f-RevGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1606.00709) | \u003Cimg src = 'viz\u002FfGAN\u002Freverse_kl\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FfGAN\u002Freverse_kl\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FfGAN_reverse_kl_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FfGAN_reverse_kl_loss.png' height = '150px'>\n[f-HellingerGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1606.00709) | \u003Cimg src = 'viz\u002FfGAN\u002Fhellinger\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FfGAN\u002Fhellinger\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FfGAN_hellinger_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FfGAN_hellinger_loss.png' height = '150px'>\n[VAE](https:\u002F\u002Farxiv.org\u002Fabs\u002F1312.6114) | \u003Cimg src = 'viz\u002FVAE\u002Fsample_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FVAE\u002Fsample_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FVAE_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FVAE_loss.png' height = '150px'>\n[BIRVAE](https:\u002F\u002Farxiv.org\u002Fabs\u002F1807.07306) | \u003Cimg src = 'viz\u002FBIRVAE\u002Fsample_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FBIRVAE\u002Fsample_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FBIRVAE_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FBIRVAE_loss.png' height = '150px'>\n\n# To Do\nModels: CVAE, denoising VAE, adversarial autoencoder | Bayesian GAN, Self-attention GAN, Primal-Dual Wasserstein GAN  \nArchitectures: Add DCGAN option  \nDatasets: Beyond MNIST\n","# 概述\nPyTorch 0.4.1 | Python 3.6.5\n\n提供了针对以下模型的注释实现及对比介绍：极大极小博弈、非饱和式、Wasserstein、带梯度惩罚的Wasserstein、最小二乘法、深度后悔分析、有界均衡、相对论式、f-散度、Fisher散度以及信息生成对抗网络（GAN），同时还包括标准变分、变分有界信息率自编码器（VAE）。\n\n每个文件开头都附有论文链接及简要摘要。可在`src`文件夹中找到可通过终端运行的脚本，或在`notebooks`文件夹中通过本地浏览器查看Jupyter Notebook可视化内容。主要的代码改动可见于`Trainer`类中的[```train```](https:\u002F\u002Fgithub.com\u002Fshayneobrien\u002Fgenerative-models\u002Fblob\u002Fmaster\u002Fsrc\u002Fns_gan.py#L94-L170)、[```train_D```](https:\u002F\u002Fgithub.com\u002Fshayneobrien\u002Fgenerative-models\u002Fblob\u002Fmaster\u002Fsrc\u002Fns_gan.py#L172-L194)和[```train_G```](https:\u002F\u002Fgithub.com\u002Fshayneobrien\u002Fgenerative-models\u002Fblob\u002Fmaster\u002Fsrc\u002Fns_gan.py#L196-L216)，不过改动并不局限于这些部分（例如，Wasserstein GAN会在`train`函数中对权重进行裁剪，BEGAN则从`train_D`中输出多个结果，而fGAN在`viz_loss`函数中稍作修改，以便在标题中注明所用方法）。\n\n本仓库中的所有代码均以生成式、无监督的方式处理二值（黑白）MNIST数据集。其架构兼容多种数据类型（一维、二维以及方形三维图像）。绘图函数可支持二值与RGB图像。若检测到GPU，则模型会优先使用GPU；否则将默认使用CPU。VAE的训练器类包含用于可视化潜在空间表示的方法（参见[```make_all```](https:\u002F\u002Fgithub.com\u002Fshayneobrien\u002Fgenerative-models\u002Fblob\u002Fmaster\u002Fsrc\u002Fvae.py#L333-L343)函数）。\n\n# 使用方法\n初始化环境：\n```\npython -m venv env  \n. env\u002Fbin\u002Factivate  \npip install -r requirements.txt  \n```\n\n若需在Jupyter Notebook中进行实验：\n```\njupyter notebook\n```\n\n若需通过终端运行：\n```\ncd src\npython bir_vae.py\n```\n\n# 新模型\n本仓库的主要目的之一是尽可能简化深度生成模型（即GAN\u002FVAE）变体的实现。之所以能做到这一点，是因为通常情况下（但并非总是如此，如BIRVAE），提出的改进仅涉及反向传播时损失的计算方式。因此，核心训练类的设计使得大多数新模型的实现只需修改GAN训练器类中的`train_D`和`train_G`函数，以及VAE训练器类中的`compute_batch`函数即可。\n\n假设我们有一个非饱和式GAN，并希望将其改造成最小二乘法GAN。为此，我们只需更改两行代码：\n\n【原版】（NSGAN）\n```python\ndef train_D(self, images):\n  ...\n  D_loss = -torch.mean(torch.log(DX_score + 1e-8) + torch.log(1 - DG_score + 1e-8))\n\n  return D_loss\n```\n```python\ndef train_G(self, images):\n  ...\n  G_loss = -torch.mean(torch.log(DG_score + 1e-8))\n\n  return G_loss\n```\n\n【新版】（LSGAN）\n```python\ndef train_D(self, images):\n  ...\n  D_loss = (0.50 * torch.mean((DX_score - 1.)**2)) + (0.50 * torch.mean((DG_score - 0.)**2))\n\n  return D_loss\n```\n```python\ndef train_G(self, images):\n  ...\n  G_loss = 0.50 * torch.mean((DG_score - 1.)**2)\n\n  return G_loss\n```\n\n# 模型架构\n本实现中，生成器（G）和判别器（D）均采用简单的两层前馈神经网络架构。虽然该架构对于MNIST数据集能够产生合理的结果，但在实际应用中，建议使用深层卷积架构（如DCGAN）以获得更高质量的输出。这可以通过修改GAN的生成器和判别器类，或VAE的编码器和解码器类来实现。\n\n# 可视化\n所有模型均训练了25个 epoch，隐藏层维度为400，潜在空间维度为20。其他实现细节尽可能与各自原始论文（已链接）保持一致。\n\n*模型* | *第1轮* | *第25轮* | *进度* | *损失*\n:---: | :---: | :---: | :---: | :---: |\n[MMGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1406.2661) | \u003Cimg src = 'viz\u002FMMGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FMMGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FMMGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FMMGAN_loss.png' height = '150px'>\n[NSGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1406.2661) | \u003Cimg src = 'viz\u002FNSGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FNSGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FNSGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FNSGAN_loss.png' height = '150px'>\n[WGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1701.07875) | \u003Cimg src = 'viz\u002FWGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FWGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FWGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FWGAN_loss.png' height = '150px'>\n[WGPGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1704.00028) | \u003Cimg src = 'viz\u002FWGPGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FWGPGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FWGPGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FWGPGAN_loss.png' height = '150px'>\n[DRAGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1705.07215) | \u003Cimg src = 'viz\u002FDRAGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FDRAGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FDRAGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FDRAGAN_loss.png' height = '150px'>\n[BEGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1703.10717) | \u003Cimg src = 'viz\u002FBEGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FBEGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FBEGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FBEGAN_loss.png' height = '150px'>\n[LSGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1611.04076) | \u003Cimg src = 'viz\u002FLSGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FLSGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FLSGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FLSGAN_loss.png' height = '150px'>\n[RaNSGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1807.00734) | \u003Cimg src = 'viz\u002FRaNSGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FRaNSGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FRaNSGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FRaNSGAN_loss.png' height = '150px'>\n[FisherGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1606.07536) | \u003Cimg src = 'viz\u002FFisherGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FFisherGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FFisherGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FFisherGAN_loss.png' height = '150px'>\n[InfoGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1606.03657) | \u003Cimg src = 'viz\u002FInfoGAN\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FInfoGAN\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FInfoGAN_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FInfoGAN_loss.png' height = '150px'>\n[f-TVGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1606.00709) | \u003Cimg src = 'viz\u002FfGAN\u002Ftotal_variation\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FfGAN\u002Ftotal_variation\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FfGAN_total_variation_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FfGAN_total_variation_loss.png' height = '150px'>\n[f-PearsonGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1606.00709) | \u003Cimg src = 'viz\u002FfGAN\u002Fpearson\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FfGAN\u002Fpearson\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FfGAN_pearson_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FfGAN_pearson_loss.png' height = '150px'>\n[f-JSGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1606.00709) | \u003Cimg src = 'viz\u002FfGAN\u002Fjensen_shannon\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FfGAN\u002Fjensen_shannon\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FfGAN_jensen_shannon_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FfGAN_jensen_shannon_loss.png' height = '150px'>\n[f-ForwGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1606.00709) | \u003Cimg src = 'viz\u002FfGAN\u002Fforward_kl\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FfGAN\u002Fforward_kl\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FfGAN_forward_kl_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FfGAN_forward_kl_loss.png' height = '150px'>\n[f-RevGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1606.00709) | \u003Cimg src = 'viz\u002FfGAN\u002Freverse_kl\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FfGAN\u002Freverse_kl\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FfGAN_reverse_kl_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FfGAN_reverse_kl_loss.png' height = '150px'>\n[f-HellingerGAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F1606.00709) | \u003Cimg src = 'viz\u002FfGAN\u002Fhellinger\u002Freconst_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FfGAN\u002Fhellinger\u002Freconst_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FfGAN_hellinger_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FfGAN_hellinger_loss.png' height = '150px'>\n[VAE](https:\u002F\u002Farxiv.org\u002Fabs\u002F1312.6114) | \u003Cimg src = 'viz\u002FVAE\u002Fsample_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FVAE\u002Fsample_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FVAE_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FVAE_loss.png' height = '150px'>\n[BIRVAE](https:\u002F\u002Farxiv.org\u002Fabs\u002F1807.07306) | \u003Cimg src = 'viz\u002FBIRVAE\u002Fsample_1.png' height = '150px'> | \u003Cimg src = 'viz\u002FBIRVAE\u002Fsample_25.png' height = '150px'> | \u003Cimg src = 'viz\u002Fgifs\u002FBIRVAE_gif.gif' height = '150px'> | \u003Cimg src = 'viz\u002Flosses\u002FBIRVAE_loss.png' height = '150px'>\n\n# 待办事项\n模型：CVAE、去噪 VAE、对抗自编码器 | 贝叶斯 GAN、自注意力 GAN、原对偶 Wasserstein GAN  \n架构：增加 DCGAN 选项  \n数据集：超越 MNIST","# generative-models 快速上手指南\n\n本仓库提供了多种生成对抗网络（GAN）和变分自编码器（VAE）变体的注释实现，包括 Minimax、Wasserstein、Least Squares、InfoGAN 等。代码基于 PyTorch，默认在二值化 MNIST 数据集上以无监督方式运行。\n\n## 环境准备\n\n*   **操作系统**: Linux \u002F macOS \u002F Windows\n*   **Python 版本**: 推荐 Python 3.6.5 或更高版本\n*   **核心依赖**: PyTorch 0.4.1 或更高版本\n*   **硬件要求**: 支持 CPU 运行；若检测到 GPU (CUDA)，将自动使用 GPU 加速。\n\n## 安装步骤\n\n1.  **创建虚拟环境**\n    在项目根目录下执行以下命令创建隔离环境：\n    ```bash\n    python -m venv env\n    ```\n\n2.  **激活环境**\n    *   **Linux\u002FmacOS**:\n        ```bash\n        . env\u002Fbin\u002Factivate\n        ```\n    *   **Windows**:\n        ```bash\n        env\\Scripts\\activate\n        ```\n\n3.  **安装依赖**\n    安装项目所需的 Python 包。\n    > **提示**：国内用户建议使用清华或阿里镜像源加速安装。\n    ```bash\n    pip install -r requirements.txt -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n    ```\n\n## 基本使用\n\n### 方式一：终端直接运行（推荐）\n\n这是最简单的运行方式，直接在命令行启动训练脚本。以下以运行边界信息率变分自编码器（BIR-VAE）为例：\n\n```bash\ncd src\npython bir_vae.py\n```\n\n*   程序会自动加载二值化 MNIST 数据集。\n*   若本地有 GPU，模型将自动使用 GPU 训练，否则回退至 CPU。\n*   训练过程及结果可视化将在运行时生成。\n\n### 方式二：Jupyter Notebook 交互探索\n\n如果你希望通过 Notebook 查看可视化结果或逐步调试：\n\n1.  确保已激活虚拟环境。\n2.  启动 Jupyter 服务：\n    ```bash\n    jupyter notebook\n    ```\n3.  在浏览器中打开 `notebooks` 文件夹下的对应文件进行实验。\n\n### 自定义模型架构\n\n默认实现使用了简单的两层前馈网络以适应 MNIST。若需获得更高质量的图像生成效果（如处理 RGB 图片），建议修改 `src` 目录下的代码：\n*   **GAN 模型**: 编辑 `Generator` 和 `Discriminator` 类，替换为深度卷积架构（如 DCGAN）。\n*   **VAE 模型**: 编辑 `Encoder` 和 `Decoder` 类。\n\n大多数新模型的实现只需修改 Trainer 类中的损失计算函数（如 `train_D`, `train_G` 或 `compute_batch`），无需重构整个训练流程。","某高校计算机视觉实验室的研究团队正在复现多篇关于生成对抗网络（GAN）的顶会论文，旨在对比不同损失函数对图像生成稳定性的影响。\n\n### 没有 generative-models 时\n- **代码复用率低**：每尝试一种新模型（如从 NSGAN 切换到 LSGAN），研究人员需从头重写判别器和生成器的训练循环，大量重复劳动导致开发效率低下。\n- **理论落地困难**：论文中的数学公式转化为 PyTorch 代码时容易出错，尤其是复杂的梯度惩罚或相对论平均项，调试过程耗时且难以定位逻辑偏差。\n- **缺乏直观对比**：不同模型的潜在空间分布和生成效果分散在独立的脚本中，缺乏统一的可视化接口，难以在同一基准下直观评估各算法优劣。\n- **环境配置繁琐**：每次运行新实验都需手动调整数据加载、GPU 检测及超参数设置，极易因配置疏漏导致实验结果不可复现。\n\n### 使用 generative-models 后\n- **极速模型切换**：借助其模块化设计，团队仅需修改 `train_D` 和 `train_G` 函数中的几行损失计算代码，即可在几分钟内将非饱和 GAN 转换为最小二乘 GAN 或 FisherGAN。\n- **源码即文档**：每个文件头部附带论文链接与摘要，核心训练逻辑包含详细注释，研究人员可直接对照公式阅读代码，大幅降低了理解门槛和实现错误率。\n- **统一可视化分析**：利用内置的 Jupyter Notebook 和绘图函数，团队能一键生成所有模型的潜在空间漫步图和生成样本对比，快速验证假设并撰写实验报告。\n- **开箱即用体验**：预置的环境配置脚本和自动 GPU 检测机制，让团队成员无需关心底层细节，直接聚焦于算法原理的创新与验证。\n\ngenerative-models 通过将复杂的生成模型变体标准化为可插拔的代码模块，极大地缩短了从理论推导到实验验证的周期，成为科研人员进行生成式算法探索的高效加速器。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fshayneobrien_generative-models_5f345224.png","shayneobrien","Shayne O'Brien","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fshayneobrien_dcdd2a2a.jpg","NLP \u002F Machine Learning \u002F Network Science. Moved from MIT to Apple 06\u002F2019","MIT","San Francisco, CA",null,"https:\u002F\u002Fwww.linkedin.com\u002Fin\u002Fshaynetobrien\u002F","https:\u002F\u002Fgithub.com\u002Fshayneobrien",[82,86],{"name":83,"color":84,"percentage":85},"Jupyter Notebook","#DA5B0B",99,{"name":87,"color":88,"percentage":89},"Python","#3572A5",1,505,72,"2026-04-17T07:28:02","未说明","非必需。代码会自动检测 GPU，若未检测到则默认使用 CPU。未指定具体显卡型号、显存大小或 CUDA 版本。",{"notes":96,"python":97,"dependencies":98},"该工具主要基于二值化（黑白）MNIST 数据集运行。虽然架构兼容多种数据类型，但默认生成器和判别器仅为简单的两层前馈网络，建议用户自行修改为深度卷积架构（如 DCGAN）以获得更好的生成效果。环境初始化需使用 venv 并安装 requirements.txt 中的依赖。","3.6.5",[99,100],"torch==0.4.1","jupyter",[14,15],[103,104,105,106,107,108,109,110,64,111,112,113,114,115,116,117,118,119,120,121],"python","generative-adversarial-network","machine-learning","pytorch","discriminator","gan","wasserstein","vae","began","dragan","autoencoder","infogan","lsgan","mmgan","nsgan","ragan","wgan","fgan","fishergan","2026-03-27T02:49:30.150509","2026-04-20T04:08:09.864248",[],[]]