[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-krasserm--super-resolution":3,"tool-krasserm--super-resolution":64},[4,17,27,35,43,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},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。",162132,3,"2026-04-05T11:01:52",[13,14,15],"开发框架","图像","Agent","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},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",140436,2,"2026-04-05T23:32:43",[13,15,26],"语言模型",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":23,"last_commit_at":33,"category_tags":34,"status":16},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。",107662,"2026-04-03T11:11:01",[13,14,15],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":23,"last_commit_at":41,"category_tags":42,"status":16},3704,"NextChat","ChatGPTNextWeb\u002FNextChat","NextChat 是一款轻量且极速的 AI 助手，旨在为用户提供流畅、跨平台的大模型交互体验。它完美解决了用户在多设备间切换时难以保持对话连续性，以及面对众多 AI 模型不知如何统一管理的痛点。无论是日常办公、学习辅助还是创意激发，NextChat 都能让用户随时随地通过网页、iOS、Android、Windows、MacOS 或 Linux 端无缝接入智能服务。\n\n这款工具非常适合普通用户、学生、职场人士以及需要私有化部署的企业团队使用。对于开发者而言，它也提供了便捷的自托管方案，支持一键部署到 Vercel 或 Zeabur 等平台。\n\nNextChat 的核心亮点在于其广泛的模型兼容性，原生支持 Claude、DeepSeek、GPT-4 及 Gemini Pro 等主流大模型，让用户在一个界面即可自由切换不同 AI 能力。此外，它还率先支持 MCP（Model Context Protocol）协议，增强了上下文处理能力。针对企业用户，NextChat 提供专业版解决方案，具备品牌定制、细粒度权限控制、内部知识库整合及安全审计等功能，满足公司对数据隐私和个性化管理的高标准要求。",87618,"2026-04-05T07:20:52",[13,26],{"id":44,"name":45,"github_repo":46,"description_zh":47,"stars":48,"difficulty_score":23,"last_commit_at":49,"category_tags":50,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",84991,"2026-04-05T10:45:23",[14,51,52,53,15,54,26,13,55],"数据工具","视频","插件","其他","音频",{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":10,"last_commit_at":62,"category_tags":63,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[15,14,13,26,54],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":69,"readme_en":70,"readme_zh":71,"quickstart_zh":72,"use_case_zh":73,"hero_image_url":74,"owner_login":75,"owner_name":76,"owner_avatar_url":77,"owner_bio":78,"owner_company":79,"owner_location":80,"owner_email":81,"owner_twitter":81,"owner_website":82,"owner_url":83,"languages":84,"stars":89,"forks":90,"last_commit_at":91,"license":92,"difficulty_score":10,"env_os":93,"env_gpu":94,"env_ram":94,"env_deps":95,"category_tags":99,"github_topics":100,"view_count":10,"oss_zip_url":81,"oss_zip_packed_at":81,"status":16,"created_at":108,"updated_at":109,"faqs":110,"releases":140},1136,"krasserm\u002Fsuper-resolution","super-resolution","Tensorflow 2.x based implementation of EDSR, WDSR and SRGAN for single image super-resolution","super-resolution是一个基于TensorFlow 2.x的图像超分辨率工具库，专注于通过深度学习技术提升单张图片的分辨率和清晰度。它实现了EDSR、WDSR和SRGAN等获奖模型，能够将低分辨率图像转换为更细腻、更接近真实场景的高分辨率版本。工具通过自动下载DIV2K数据集并支持多种降质方式（如bicubic、unknown等），为图像增强提供了灵活的训练和测试环境。开发者可通过高阶API直接训练模型，或在SRGAN框架下对EDSR\u002FWDSR进行微调，适合需要图像质量优化的技术团队和研究者。其技术亮点包括对竞赛优胜模型的完整实现、可扩展的训练流程以及针对不同应用场景的参数配置能力。","![Travis CI](https:\u002F\u002Ftravis-ci.com\u002Fkrasserm\u002Fsuper-resolution.svg?branch=master)\n\n# Single Image Super-Resolution with EDSR, WDSR and SRGAN\n\nA [Tensorflow 2.x](https:\u002F\u002Fwww.tensorflow.org\u002Fbeta) based implementation of\n\n- [Enhanced Deep Residual Networks for Single Image Super-Resolution](https:\u002F\u002Farxiv.org\u002Fabs\u002F1707.02921) (EDSR), winner \n  of the [NTIRE 2017](http:\u002F\u002Fwww.vision.ee.ethz.ch\u002Fntire17\u002F) super-resolution challenge.\n- [Wide Activation for Efficient and Accurate Image Super-Resolution](https:\u002F\u002Farxiv.org\u002Fabs\u002F1808.08718) (WDSR), winner \n  of the [NTIRE 2018](http:\u002F\u002Fwww.vision.ee.ethz.ch\u002Fntire18\u002F) super-resolution challenge (realistic tracks).\n- [Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network](https:\u002F\u002Farxiv.org\u002Fabs\u002F1609.04802) (SRGAN).\n\nThis is a complete re-write of the old Keras\u002FTensorflow 1.x based implementation available [here](https:\u002F\u002Fgithub.com\u002Fkrasserm\u002Fsuper-resolution\u002Ftree\u002Fprevious).\nSome parts are still work in progress but you can already train models as described in the papers via a high-level training \nAPI. Furthermore, you can also [fine-tune](#srgan-for-fine-tuning-edsr-and-wdsr-models) EDSR and WDSR models in an SRGAN \ncontext. [Training](#training) and [usage](#getting-started) examples are given in the notebooks\n\n- [example-edsr.ipynb](example-edsr.ipynb)\n- [example-wdsr.ipynb](example-wdsr.ipynb)\n- [example-srgan.ipynb](example-srgan.ipynb) \n\nA `DIV2K` [data provider](#div2k-dataset) automatically downloads [DIV2K](https:\u002F\u002Fdata.vision.ee.ethz.ch\u002Fcvl\u002FDIV2K\u002F) \ntraining and validation images of given scale (2, 3, 4 or 8) and downgrade operator (\"bicubic\", \"unknown\", \"mild\" or \n\"difficult\"). \n\n**Important:** if you want to evaluate the pre-trained models with a dataset other than DIV2K please read \n[this comment](https:\u002F\u002Fgithub.com\u002Fkrasserm\u002Fsuper-resolution\u002Fissues\u002F19#issuecomment-586114933) (and replies) first.  \n\n## Environment setup\n\nCreate a new [conda](https:\u002F\u002Fconda.io) environment with\n\n    conda env create -f environment.yml\n    \nand activate it with\n\n    conda activate sisr\n\n## Introduction\n\nYou can find an introduction to single-image super-resolution in [this article](https:\u002F\u002Fkrasserm.github.io\u002F2019\u002F09\u002F04\u002Fsuper-resolution\u002F). \nIt also demonstrates how EDSR and WDSR models can be fine-tuned with SRGAN (see also [this section](#srgan-for-fine-tuning-edsr-and-wdsr-models)).\n\n## Getting started \n\nExamples in this section require following pre-trained weights for running (see also example notebooks):  \n\n### Pre-trained weights\n\n- [weights-edsr-16-x4.tar.gz](https:\u002F\u002Fmartin-krasser.de\u002Fsisr\u002Fweights-edsr-16-x4.tar.gz) \n    - EDSR x4 baseline as described in the EDSR paper: 16 residual blocks, 64 filters, 1.52M parameters. \n    - PSNR on DIV2K validation set = 28.89 dB (images 801 - 900, 6 + 4 pixel border included).\n- [weights-wdsr-b-32-x4.tar.gz](https:\u002F\u002Fmartin-krasser.de\u002Fsisr\u002Fweights-wdsr-b-32-x4.tar.gz) \n    - WDSR B x4 custom model: 32 residual blocks, 32 filters, expansion factor 6, 0.62M parameters. \n    - PSNR on DIV2K validation set = 28.91 dB (images 801 - 900, 6 + 4 pixel border included).\n- [weights-srgan.tar.gz](https:\u002F\u002Fmartin-krasser.de\u002Fsisr\u002Fweights-srgan.tar.gz) \n    - SRGAN as described in the SRGAN paper: 1.55M parameters, trained with VGG54 content loss.\n    \nAfter download, extract them in the root folder of the project with\n\n    tar xvfz weights-\u003C...>.tar.gz\n\n### EDSR\n\n```python\nfrom model import resolve_single\nfrom model.edsr import edsr\n\nfrom utils import load_image, plot_sample\n\nmodel = edsr(scale=4, num_res_blocks=16)\nmodel.load_weights('weights\u002Fedsr-16-x4\u002Fweights.h5')\n\nlr = load_image('demo\u002F0851x4-crop.png')\nsr = resolve_single(model, lr)\n\nplot_sample(lr, sr)\n```\n\n![result-edsr](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fkrasserm_super-resolution_readme_d4b809fa0936.png)\n\n### WDSR\n\n```python\nfrom model.wdsr import wdsr_b\n\nmodel = wdsr_b(scale=4, num_res_blocks=32)\nmodel.load_weights('weights\u002Fwdsr-b-32-x4\u002Fweights.h5')\n\nlr = load_image('demo\u002F0829x4-crop.png')\nsr = resolve_single(model, lr)\n\nplot_sample(lr, sr)\n```\n\n![result-wdsr](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fkrasserm_super-resolution_readme_7e1d5d9e7075.png)\n\nWeight normalization in WDSR models is implemented with the new `WeightNormalization` layer wrapper of \n[Tensorflow Addons](https:\u002F\u002Fgithub.com\u002Ftensorflow\u002Faddons). In its latest version, this wrapper seems to \ncorrupt weights when running `model.predict(...)`. A workaround is to set `model.run_eagerly = True` or \ncompile the model with `model.compile(loss='mae')` in advance. This issue doesn't arise when calling the\nmodel directly with `model(...)` though. To be further investigated ... \n\n### SRGAN\n\n```python\nfrom model.srgan import generator\n\nmodel = generator()\nmodel.load_weights('weights\u002Fsrgan\u002Fgan_generator.h5')\n\nlr = load_image('demo\u002F0869x4-crop.png')\nsr = resolve_single(model, lr)\n\nplot_sample(lr, sr)\n```\n\n![result-srgan](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fkrasserm_super-resolution_readme_008afd638104.png)\n\n## DIV2K dataset\n\nFor training and validation on [DIV2K](https:\u002F\u002Fdata.vision.ee.ethz.ch\u002Fcvl\u002FDIV2K\u002F) images, applications should use the \nprovided `DIV2K` data loader. It automatically downloads DIV2K images to `.div2k` directory and converts them to a \ndifferent format for faster loading.\n\n### Training dataset\n\n```python\nfrom data import DIV2K\n\ntrain_loader = DIV2K(scale=4,             # 2, 3, 4 or 8\n                     downgrade='bicubic', # 'bicubic', 'unknown', 'mild' or 'difficult' \n                     subset='train')      # Training dataset are images 001 - 800\n                     \n# Create a tf.data.Dataset          \ntrain_ds = train_loader.dataset(batch_size=16,         # batch size as described in the EDSR and WDSR papers\n                                random_transform=True, # random crop, flip, rotate as described in the EDSR paper\n                                repeat_count=None)     # repeat iterating over training images indefinitely\n\n# Iterate over LR\u002FHR image pairs                                \nfor lr, hr in train_ds:\n    # .... \n```\n\nCrop size in HR images is 96x96. \n\n### Validation dataset\n\n```python\nfrom data import DIV2K\n\nvalid_loader = DIV2K(scale=4,             # 2, 3, 4 or 8\n                     downgrade='bicubic', # 'bicubic', 'unknown', 'mild' or 'difficult' \n                     subset='valid')      # Validation dataset are images 801 - 900\n                     \n# Create a tf.data.Dataset          \nvalid_ds = valid_loader.dataset(batch_size=1,           # use batch size of 1 as DIV2K images have different size\n                                random_transform=False, # use DIV2K images in original size \n                                repeat_count=1)         # 1 epoch\n                                \n# Iterate over LR\u002FHR image pairs                                \nfor lr, hr in valid_ds:\n    # ....                                 \n```\n\n## Training \n\nThe following training examples use the [training and validation datasets](#div2k-dataset) described earlier. The high-level \ntraining API is designed around *steps* (= minibatch updates) rather than *epochs* to better match the descriptions in the \npapers.\n\n## EDSR\n\n```python\nfrom model.edsr import edsr\nfrom train import EdsrTrainer\n\n# Create a training context for an EDSR x4 model with 16 \n# residual blocks.\ntrainer = EdsrTrainer(model=edsr(scale=4, num_res_blocks=16), \n                      checkpoint_dir=f'.ckpt\u002Fedsr-16-x4')\n                      \n# Train EDSR model for 300,000 steps and evaluate model\n# every 1000 steps on the first 10 images of the DIV2K\n# validation set. Save a checkpoint only if evaluation\n# PSNR has improved.\ntrainer.train(train_ds,\n              valid_ds.take(10),\n              steps=300000, \n              evaluate_every=1000, \n              save_best_only=True)\n              \n# Restore from checkpoint with highest PSNR.\ntrainer.restore()\n\n# Evaluate model on full validation set.\npsnr = trainer.evaluate(valid_ds)\nprint(f'PSNR = {psnr.numpy():3f}')\n\n# Save weights to separate location.\ntrainer.model.save_weights('weights\u002Fedsr-16-x4\u002Fweights.h5')                                    \n```\n\nInterrupting training and restarting it again resumes from the latest saved checkpoint. The trained Keras model can be\naccessed with `trainer.model`.\n\n## WDSR\n\n```python\nfrom model.wdsr import wdsr_b\nfrom train import WdsrTrainer\n\n# Create a training context for a WDSR B x4 model with 32 \n# residual blocks.\ntrainer = WdsrTrainer(model=wdsr_b(scale=4, num_res_blocks=32), \n                      checkpoint_dir=f'.ckpt\u002Fwdsr-b-8-x4')\n\n# Train WDSR B model for 300,000 steps and evaluate model\n# every 1000 steps on the first 10 images of the DIV2K\n# validation set. Save a checkpoint only if evaluation\n# PSNR has improved.\ntrainer.train(train_ds,\n              valid_ds.take(10),\n              steps=300000, \n              evaluate_every=1000, \n              save_best_only=True)\n\n# Restore from checkpoint with highest PSNR.\ntrainer.restore()\n\n# Evaluate model on full validation set.\npsnr = trainer.evaluate(valid_ds)\nprint(f'PSNR = {psnr.numpy():3f}')\n\n# Save weights to separate location.\ntrainer.model.save_weights('weights\u002Fwdsr-b-32-x4\u002Fweights.h5')\n```\n\n## SRGAN\n\n### Generator pre-training\n\n```python\nfrom model.srgan import generator\nfrom train import SrganGeneratorTrainer\n\n# Create a training context for the generator (SRResNet) alone.\npre_trainer = SrganGeneratorTrainer(model=generator(), checkpoint_dir=f'.ckpt\u002Fpre_generator')\n\n# Pre-train the generator with 1,000,000 steps (100,000 works fine too). \npre_trainer.train(train_ds, valid_ds.take(10), steps=1000000, evaluate_every=1000)\n\n# Save weights of pre-trained generator (needed for fine-tuning with GAN).\npre_trainer.model.save_weights('weights\u002Fsrgan\u002Fpre_generator.h5')\n```\n\n### Generator fine-tuning (GAN)\n\n```python\nfrom model.srgan import generator, discriminator\nfrom train import SrganTrainer\n\n# Create a new generator and init it with pre-trained weights.\ngan_generator = generator()\ngan_generator.load_weights('weights\u002Fsrgan\u002Fpre_generator.h5')\n\n# Create a training context for the GAN (generator + discriminator).\ngan_trainer = SrganTrainer(generator=gan_generator, discriminator=discriminator())\n\n# Train the GAN with 200,000 steps.\ngan_trainer.train(train_ds, steps=200000)\n\n# Save weights of generator and discriminator.\ngan_trainer.generator.save_weights('weights\u002Fsrgan\u002Fgan_generator.h5')\ngan_trainer.discriminator.save_weights('weights\u002Fsrgan\u002Fgan_discriminator.h5')\n```\n\n## SRGAN for fine-tuning EDSR and WDSR models\n\nIt is also possible to fine-tune EDSR and WDSR x4 models with SRGAN. They can be used as drop-in replacement for the\noriginal SRGAN generator. More details in [this article](https:\u002F\u002Fkrasserm.github.io\u002F2019\u002F09\u002F04\u002Fsuper-resolution\u002F).\n\n```python\n# Create EDSR generator and init with pre-trained weights\ngenerator = edsr(scale=4, num_res_blocks=16)\ngenerator.load_weights('weights\u002Fedsr-16-x4\u002Fweights.h5')\n\n# Fine-tune EDSR model via SRGAN training.\ngan_trainer = SrganTrainer(generator=generator, discriminator=discriminator())\ngan_trainer.train(train_ds, steps=200000)\n```\n\n```python\n# Create WDSR B generator and init with pre-trained weights\ngenerator = wdsr_b(scale=4, num_res_blocks=32)\ngenerator.load_weights('weights\u002Fwdsr-b-16-32\u002Fweights.h5')\n\n# Fine-tune WDSR B  model via SRGAN training.\ngan_trainer = SrganTrainer(generator=generator, discriminator=discriminator())\ngan_trainer.train(train_ds, steps=200000)\n```\n","![Travis CI](https:\u002F\u002Ftravis-ci.com\u002Fkrasserm\u002Fsuper-resolution.svg?branch=master)\n\n# 使用EDSR、WDSR和SRGAN的单图像超分辨率\n\n基于[TensorFlow 2.x](https:\u002F\u002Fwww.tensorflow.org\u002Fbeta)的实现，包括：\n\n- [用于单图像超分辨率的增强深度残差网络](https:\u002F\u002Farxiv.org\u002Fabs\u002F1707.02921)（EDSR），该模型赢得了[NTIRE 2017](http:\u002F\u002Fwww.vision.ee.ethz.ch\u002Fntire17\u002F)超分辨率挑战赛。\n- [用于高效且精确图像超分辨率的宽激活函数](https:\u002F\u002Farxiv.org\u002Fabs\u002F1808.08718)（WDSR），该模型赢得了[NTIRE 2018](http:\u002F\u002Fwww.vision.ee.ethz.ch\u002Fntire18\u002F)超分辨率挑战赛（真实场景赛道）。\n- [使用生成对抗网络实现照片级真实的单图像超分辨率](https:\u002F\u002Farxiv.org\u002Fabs\u002F1609.04802)（SRGAN）。\n\n这是对之前基于Keras\u002FTensorFlow 1.x的实现的全新重写，旧版本可在[这里](https:\u002F\u002Fgithub.com\u002Fkrasserm\u002Fsuper-resolution\u002Ftree\u002Fprevious)找到。部分功能仍在开发中，但您已经可以通过高级训练API按照论文中的方法训练模型。此外，您还可以在SRGAN框架下对EDSR和WDSR模型进行[微调](#srgan-for-fine-tuning-edsr-and-wdsr-models)。训练和使用示例分别在以下notebook中给出：\n\n- [example-edsr.ipynb](example-edsr.ipynb)\n- [example-wdsr.ipynb](example-wdsr.ipynb)\n- [example-srgan.ipynb](example-srgan.ipynb)\n\n一个`DIV2K`数据提供者会自动下载指定缩放倍数（2、3、4或8）和降质算子（“双三次”、“未知”、“轻微”或“困难”）的`DIV2K`训练和验证图像。\n\n**重要提示：** 如果您希望使用除`DIV2K`之外的数据集来评估预训练模型，请先阅读[此评论](https:\u002F\u002Fgithub.com\u002Fkrasserm\u002Fsuper-resolution\u002Fissues\u002F19#issuecomment-586114933)及其回复。\n\n## 环境设置\n\n创建一个新的[conda](https:\u002F\u002Fconda.io)环境：\n\n    conda env create -f environment.yml\n    \n然后激活它：\n\n    conda activate sisr\n\n## 简介\n\n您可以在[这篇文章](https:\u002F\u002Fkrasserm.github.io\u002F2019\u002F09\u002F04\u002Fsuper-resolution\u002F)中找到关于单图像超分辨率的介绍。文中还演示了如何使用SRGAN对EDSR和WDSR模型进行微调（参见[本节](#srgan-for-fine-tuning-edsr-and-wdsr-models)）。\n\n## 快速入门\n\n本节中的示例需要以下预训练权重才能运行（请参阅示例notebook）：\n\n### 预训练权重\n\n- [weights-edsr-16-x4.tar.gz](https:\u002F\u002Fmartin-krasser.de\u002Fsisr\u002Fweights-edsr-16-x4.tar.gz)\n    - EDSR x4基准模型，如EDSR论文所述：16个残差块，64个滤波器，152万参数。\n    - 在`DIV2K`验证集上的PSNR为28.89 dB（图像801–900，包含6+4像素边框）。\n- [weights-wdsr-b-32-x4.tar.gz](https:\u002F\u002Fmartin-krasser.de\u002Fsisr\u002Fweights-wdsr-b-32-x4.tar.gz)\n    - WDSR B x4自定义模型：32个残差块，32个滤波器，扩张因子6，62万参数。\n    - 在`DIV2K`验证集上的PSNR为28.91 dB（图像801–900，包含6+4像素边框）。\n- [weights-srgan.tar.gz](https:\u002F\u002Fmartin-krasser.de\u002Fsisr\u002Fweights-srgan.tar.gz)\n    - SRGAN模型，如SRGAN论文所述：155万参数，使用VGG54内容损失进行训练。\n\n下载后，将其解压到项目根目录：\n\n    tar xvfz weights-\u003C...>.tar.gz\n\n### EDSR\n\n```python\nfrom model import resolve_single\nfrom model.edsr import edsr\n\nfrom utils import load_image, plot_sample\n\nmodel = edsr(scale=4, num_res_blocks=16)\nmodel.load_weights('weights\u002Fedsr-16-x4\u002Fweights.h5')\n\nlr = load_image('demo\u002F0851x4-crop.png')\nsr = resolve_single(model, lr)\n\nplot_sample(lr, sr)\n```\n\n![result-edsr](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fkrasserm_super-resolution_readme_d4b809fa0936.png)\n\n### WDSR\n\n```python\nfrom model.wdsr import wdsr_b\n\nmodel = wdsr_b(scale=4, num_res_blocks=32)\nmodel.load_weights('weights\u002Fwdsr-b-32-x4\u002Fweights.h5')\n\nlr = load_image('demo\u002F0829x4-crop.png')\nsr = resolve_single(model, lr)\n\nplot_sample(lr, sr)\n```\n\n![result-wdsr](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fkrasserm_super-resolution_readme_7e1d5d9e7075.png)\n\nWDSR模型中的权重归一化是通过[TensorFlow Addons](https:\u002F\u002Fgithub.com\u002Ftensorflow\u002Faddons)的新`WeightNormalization`层包装实现的。在其最新版本中，当调用`model.predict(...)`时，该包装似乎会导致权重损坏。一种解决方法是将`model.run_eagerly = True`，或者提前使用`model.compile(loss='mae')`编译模型。不过，直接调用`model(...)`时并不会出现此问题。这一问题仍需进一步研究……\n\n### SRGAN\n\n```python\nfrom model.srgan import generator\n\nmodel = generator()\nmodel.load_weights('weights\u002Fsrgan\u002Fgan_generator.h5')\n\nlr = load_image('demo\u002F0869x4-crop.png')\nsr = resolve_single(model, lr)\n\nplot_sample(lr, sr)\n```\n\n![result-srgan](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fkrasserm_super-resolution_readme_008afd638104.png)\n\n## DIV2K数据集\n\n为了在`DIV2K`（[https:\u002F\u002Fdata.vision.ee.ethz.ch\u002Fcvl\u002FDIV2K\u002F](https:\u002F\u002Fdata.vision.ee.ethz.ch\u002Fcvl\u002FDIV2K\u002F)）图像上进行训练和验证，应用程序应使用提供的`DIV2K`数据加载器。它会自动将`DIV2K`图像下载到`.div2k`目录，并将其转换为更快加载的格式。\n\n### 训练数据集\n\n```python\nfrom data import DIV2K\n\ntrain_loader = DIV2K(scale=4,             # 2、3、4或8\n                     downgrade='bicubic', # 'bicubic'、'unknown'、'mild'或'difficult'\n                     subset='train')      # 训练数据集为图像001–800\n                     \n# 创建一个tf.data.Dataset          \ntrain_ds = train_loader.dataset(batch_size=16,         # 批量大小如EDSR和WDSR论文所述\n                                random_transform=True, # 如EDSR论文所述的随机裁剪、翻转、旋转\n                                repeat_count=None)     # 无限循环迭代训练图像\n\n# 遍历低分辨率\u002F高分辨率图像对                                \nfor lr, hr in train_ds:\n    # .... \n```\n\n高分辨率图像的裁剪尺寸为96×96。\n\n### 验证数据集\n\n```python\nfrom data import DIV2K\n\nvalid_loader = DIV2K(scale=4,             # 2、3、4或8\n                     downgrade='bicubic', # 'bicubic'、'unknown'、'mild'或'difficult'\n                     subset='valid')      # 验证数据集为图像801–900\n                     \n# 创建一个tf.data.Dataset          \nvalid_ds = valid_loader.dataset(batch_size=1,           # 使用批量大小为1，因为`DIV2K`图像尺寸不同\n                                random_transform=False, # 使用原始尺寸的`DIV2K`图像\n                                repeat_count=1)         # 1个epoch\n                                \n# 遍历低分辨率\u002F高分辨率图像对                                \nfor lr, hr in valid_ds:\n    # ....                                 \n```\n\n## 训练\n\n以下训练示例使用前面描述的[训练和验证数据集](#div2k-dataset)。高级训练API以*步数*（即小批量更新）而非*轮次*为核心设计，以便更好地匹配论文中的描述。\n\n## EDSR\n\n```python\nfrom model.edsr import edsr\nfrom train import EdsrTrainer\n\n# 创建一个用于训练 EDSR x4 模型的上下文，该模型包含 16 个残差块。\ntrainer = EdsrTrainer(model=edsr(scale=4, num_res_blocks=16), \n                      checkpoint_dir=f'.ckpt\u002Fedsr-16-x4')\n                      \n# 训练 EDSR 模型 300,000 步，并在 DIV2K 验证集的前 10 张图像上每 1000 步评估一次。仅当评估 PSNR 提升时才保存检查点。\ntrainer.train(train_ds,\n              valid_ds.take(10),\n              steps=300000, \n              evaluate_every=1000, \n              save_best_only=True)\n              \n# 从 PSNR 最高的检查点恢复模型。\ntrainer.restore()\n\n# 在完整的验证集上评估模型。\npsnr = trainer.evaluate(valid_ds)\nprint(f'PSNR = {psnr.numpy():3f}')\n\n# 将权重保存到单独的位置。\ntrainer.model.save_weights('weights\u002Fedsr-16-x4\u002Fweights.h5')                                    \n```\n\n中断训练并重新启动时，将从最近保存的检查点继续。训练好的 Keras 模型可以通过 `trainer.model` 访问。\n\n## WDSR\n\n```python\nfrom model.wdsr import wdsr_b\nfrom train import WdsrTrainer\n\n# 创建一个用于训练 WDSR B x4 模型的上下文，该模型包含 32 个残差块。\ntrainer = WdsrTrainer(model=wdsr_b(scale=4, num_res_blocks=32), \n                      checkpoint_dir=f'.ckpt\u002Fwdsr-b-8-x4')\n\n# 训练 WDSR B 模型 300,000 步，并在 DIV2K 验证集的前 10 张图像上每 1000 步评估一次。仅当评估 PSNR 提升时才保存检查点。\ntrainer.train(train_ds,\n              valid_ds.take(10),\n              steps=300000, \n              evaluate_every=1000, \n              save_best_only=True)\n\n# 从 PSNR 最高的检查点恢复模型。\ntrainer.restore()\n\n# 在完整的验证集上评估模型。\npsnr = trainer.evaluate(valid_ds)\nprint(f'PSNR = {psnr.numpy():3f}')\n\n# 将权重保存到单独的位置。\ntrainer.model.save_weights('weights\u002Fwdsr-b-32-x4\u002Fweights.h5')\n```\n\n## SRGAN\n\n### 生成器预训练\n\n```python\nfrom model.srgan import generator\nfrom train import SrganGeneratorTrainer\n\n# 创建一个仅针对生成器（SRResNet）的训练上下文。\npre_trainer = SrganGeneratorTrainer(model=generator(), checkpoint_dir=f'.ckpt\u002Fpre_generator')\n\n# 对生成器进行预训练，共 1,000,000 步（100,000 步也足够）。 \npre_trainer.train(train_ds, valid_ds.take(10), steps=1000000, evaluate_every=1000)\n\n# 保存预训练生成器的权重（用于后续与 GAN 的微调）。\npre_trainer.model.save_weights('weights\u002Fsrgan\u002Fpre_generator.h5')\n```\n\n### 生成器微调（GAN）\n\n```python\nfrom model.srgan import generator, discriminator\nfrom train import SrganTrainer\n\n# 创建一个新的生成器，并用预训练权重初始化。\ngan_generator = generator()\ngan_generator.load_weights('weights\u002Fsrgan\u002Fpre_generator.h5')\n\n# 创建一个针对 GAN（生成器 + 判别器）的训练上下文。\ngan_trainer = SrganTrainer(generator=gan_generator, discriminator=discriminator())\n\n# 训练 GAN 共 200,000 步。\ngan_trainer.train(train_ds, steps=200000)\n\n# 保存生成器和判别器的权重。\ngan_trainer.generator.save_weights('weights\u002Fsrgan\u002Fgan_generator.h5')\ngan_trainer.discriminator.save_weights('weights\u002Fsrgan\u002Fgan_discriminator.h5')\n```\n\n## 使用 SRGAN 微调 EDSR 和 WDSR 模型\n\n也可以使用 SRGAN 对 EDSR 和 WDSR x4 模型进行微调。这些模型可以直接替换原始的 SRGAN 生成器。更多细节请参阅[这篇文章](https:\u002F\u002Fkrasserm.github.io\u002F2019\u002F09\u002F04\u002Fsuper-resolution\u002F)。\n\n```python\n# 创建 EDSR 生成器，并用预训练权重初始化\ngenerator = edsr(scale=4, num_res_blocks=16)\ngenerator.load_weights('weights\u002Fedsr-16-x4\u002Fweights.h5')\n\n# 通过 SRGAN 训练对 EDSR 模型进行微调。\ngan_trainer = SrganTrainer(generator=generator, discriminator=discriminator())\ngan_trainer.train(train_ds, steps=200000)\n```\n\n```python\n# 创建 WDSR B 生成器，并用预训练权重初始化\ngenerator = wdsr_b(scale=4, num_res_blocks=32)\ngenerator.load_weights('weights\u002Fwdsr-b-16-32\u002Fweights.h5')\n\n# 通过 SRGAN 训练对 WDSR B 模型进行微调。\ngan_trainer = SrganTrainer(generator=generator, discriminator=discriminator())\ngan_trainer.train(train_ds, steps=200000)\n```","# Super-Resolution 快速上手指南\n\n## 环境准备\n- 系统要求：Linux\u002FmacOS (Windows需额外配置)\n- 前置依赖：\n  - Python 3.6-3.9\n  - [Conda](https:\u002F\u002Fconda.io) 环境管理器\n  - TensorFlow 2.x (通过环境配置自动安装)\n\n> 推荐使用清华镜像加速依赖安装：\n```bash\nconda config --add channels https:\u002F\u002Fmirrors.tuna.tsinghua.edu.cn\u002Fanaconda\u002Fpkgs\u002Ffree\u002F\nconda config --add channels https:\u002F\u002Fmirrors.tuna.tsinghua.edu.cn\u002Fanaconda\u002Fpkgs\u002Fmain\u002F\n```\n\n## 安装步骤\n1. 创建并激活环境：\n   ```bash\n   conda env create -f environment.yml\n   conda activate sisr\n   ```\n\n2. 下载预训练权重：\n   - EDSR: [weights-edsr-16-x4.tar.gz](https:\u002F\u002Fmartin-krasser.de\u002Fsisr\u002Fweights-edsr-16-x4.tar.gz)\n   - WDSR: [weights-wdsr-b-32-x4.tar.gz](https:\u002F\u002Fmartin-krasser.de\u002Fsisr\u002Fweights-wdsr-b-32-x4.tar.gz)\n   - SRGAN: [weights-srgan.tar.gz](https:\u002F\u002Fmartin-krasser.de\u002Fsisr\u002Fweights-srgan.tar.gz)\n\n3. 解压权重文件到项目根目录：\n   ```bash\n   tar xvfz weights-\u003Cmodel>.tar.gz\n   ```\n\n## 基本使用\n### EDSR 超分示例\n```python\nfrom model import resolve_single\nfrom model.edsr import edsr\nfrom utils import load_image, plot_sample\n\n# 加载预训练模型\nmodel = edsr(scale=4, num_res_blocks=16)\nmodel.load_weights('weights\u002Fedsr-16-x4\u002Fweights.h5')\n\n# 执行超分\nlr = load_image('demo\u002F0851x4-crop.png')\nsr = resolve_single(model, lr)\nplot_sample(lr, sr)\n```\n\n### WDSR 超分示例\n```python\nfrom model.wdsr import wdsr_b\nfrom utils import load_image, plot_sample\n\nmodel = wdsr_b(scale=4, num_res_blocks=32)\nmodel.load_weights('weights\u002Fwdsr-b-32-x4\u002Fweights.h5')\n\nlr = load_image('demo\u002F0829x4-crop.png')\nsr = resolve_single(model, lr)\nplot_sample(lr, sr)\n```\n\n### SRGAN 超分示例\n```python\nfrom model.srgan import generator\nfrom utils import load_image, plot_sample\n\nmodel = generator()\nmodel.load_weights('weights\u002Fsrgan\u002Fgan_generator.h5')\n\nlr = load_image('demo\u002F0869x4-crop.png')\nsr = resolve_single(model, lr)\nplot_sample(lr, sr)\n```","某影视特效公司需修复1990年代电视剧的低分辨率画面，以适配4K播放平台。\n\n### 没有 super-resolution 时  \n- 低分辨率画面导致人物面部纹路、服装细节模糊不清，影响观感  \n- 依赖人工逐帧手绘修复，单集耗时超200小时，成本高昂  \n- 常规算法生成的高清画面出现明显伪影（如边缘模糊、色彩失真）  \n- 无法保留原片胶片质感，画面过度锐利或失真  \n- 处理效率低下，难以满足多集同时期修复需求  \n\n### 使用 super-resolution 后  \n- EDSR模型显著增强细节清晰度，人物发丝、道具纹理等微小元素可辨  \n- 自动化处理使单集修复时间缩短至8小时，人力投入减少70%  \n- WDSR算法有效抑制伪影，画面边缘保持自然过渡，色彩还原度提升  \n- SRGAN生成画面保留原片复古色调与胶片颗粒感，符合历史还原需求  \n- 支持批量处理，同时修复5部剧集的效率提升5倍  \n\n核心价值：通过高精度超分辨率技术，在保持画面艺术风格的同时，实现高效、高质量的老旧影像数字化升级。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fkrasserm_super-resolution_d4b809fa.png","krasserm","Martin Krasser","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fkrasserm_7c2d3c95.png","Freelance ML and AI engineer","Gradion AI","Vienna, Austria",null,"https:\u002F\u002Fgradion.ai\u002F","https:\u002F\u002Fgithub.com\u002Fkrasserm",[85],{"name":86,"color":87,"percentage":88},"Python","#3572A5",100,1514,345,"2026-04-02T10:00:45","Apache-2.0","Linux, macOS, Windows","未说明",{"notes":96,"python":94,"dependencies":97},"需通过conda创建环境，预训练模型文件约5GB，WDSR需安装tensorflow-addons",[98],"TensorFlow>=2.0",[13,14],[101,102,103,67,104,105,106,107],"wdsr","edsr","keras","single-image-super-resolution","srgan","tensorflow","tensorflow2","2026-03-27T02:49:30.150509","2026-04-06T09:45:05.317027",[111,116,121,126,131,136],{"id":112,"question_zh":113,"answer_zh":114,"source_url":115},5126,"如何解决'Cannot create group in read only mode'错误？","修改callbacks.py中的model_checkpoint_after参数，仅保存权重而非整个模型。确保使用model.save()保存包含架构和权重的完整模型。","https:\u002F\u002Fgithub.com\u002Fkrasserm\u002Fsuper-resolution\u002Fissues\u002F13",{"id":117,"question_zh":118,"answer_zh":119,"source_url":120},5127,"如何解决TensorFlow 2.0兼容性问题？","在Colab中运行时，添加`%tensorflow_version 2.x`启用TensorFlow 2.0。确保代码兼容TensorFlow 2.0的API变更。","https:\u002F\u002Fgithub.com\u002Fkrasserm\u002Fsuper-resolution\u002Fissues\u002F35",{"id":122,"question_zh":123,"answer_zh":124,"source_url":125},5128,"如何优化大图像推理时的内存不足问题？","采用分块（chunking）技术处理大尺寸图像。将图像分割为小块进行推理，最后拼接结果以避免内存溢出。","https:\u002F\u002Fgithub.com\u002Fkrasserm\u002Fsuper-resolution\u002Fissues\u002F32",{"id":127,"question_zh":128,"answer_zh":129,"source_url":130},5125,"如何解决加载重新训练模型时出现的TypeError？","需要将Keras版本升级到2.2.4。在requirements.txt中添加keras=2.2.4，或通过pip安装指定版本。","https:\u002F\u002Fgithub.com\u002Fkrasserm\u002Fsuper-resolution\u002Fissues\u002F4",{"id":132,"question_zh":133,"answer_zh":134,"source_url":135},5129,"如何解决Discriminator损失值异常问题？","检查是否正确实现标签平滑和训练循环。确保生成器和判别器的损失计算逻辑符合原始论文设计，必要时调整学习率或网络结构。","https:\u002F\u002Fgithub.com\u002Fkrasserm\u002Fsuper-resolution\u002Fissues\u002F45",{"id":137,"question_zh":138,"answer_zh":139,"source_url":125},5130,"如何训练x8倍超分辨率模型？","修改edsr.py中的upsampling层结构，增加网络深度或调整超参数（如学习率、批次大小）。建议参考EDSR论文的配置进行迭代优化。",[]]