{"users":[{"id":1,"username":"lzhengning","name":null,"avatar_template":"/user_avatar/discuss.jittor.cn/lzhengning/{size}/3_2.png","flair_name":null,"admin":true,"trust_level":2},{"id":463,"username":"Yanami","name":"xhb","avatar_template":"/user_avatar/discuss.jittor.cn/yanami/{size}/766_2.png","flair_name":null,"trust_level":0},{"id":516,"username":"moonClll","name":"moonClll","avatar_template":"/user_avatar/discuss.jittor.cn/moonclll/{size}/806_2.png","flair_name":null,"trust_level":0},{"id":179,"username":"cuijianzhu","name":"Cuijianzhu","avatar_template":"/user_avatar/discuss.jittor.cn/cuijianzhu/{size}/516_2.png","flair_name":null,"trust_level":0},{"id":150,"username":"我是一个jittor用户","name":"jittor","avatar_template":"/letter_avatar_proxy/v4/letter/%E6%88%91/46a35a/{size}.png","flair_name":null,"trust_level":1},{"id":-1,"username":"system","name":"system","avatar_template":"/uploads/default/original/1X/b65920753731363feaf9a830f3c50e00352d8745.png","flair_name":null,"admin":true,"moderator":true,"trust_level":4},{"id":25,"username":"yyz","name":"zack","avatar_template":"/letter_avatar_proxy/v4/letter/y/b38774/{size}.png","flair_name":null,"trust_level":1},{"id":2,"username":"cjld","name":"Dun","avatar_template":"/user_avatar/discuss.jittor.cn/cjld/{size}/8_2.png","flair_name":null,"admin":true,"moderator":true,"trust_level":4},{"id":9,"username":"Ludong","name":"","avatar_template":"/user_avatar/discuss.jittor.cn/ludong/{size}/220_2.png","flair_name":null,"trust_level":1},{"id":8,"username":"MenghaoGuo","name":"MenghaoGuo","avatar_template":"/user_avatar/discuss.jittor.cn/menghaoguo/{size}/19_2.png","flair_name":null,"moderator":true,"trust_level":1}],"primary_groups":[],"flair_groups":[],"topic_list":{"can_create_topic":false,"more_topics_url":"/c/6-category/6?page=1","per_page":30,"top_tags":["research","模型库","official","jdet","detection","jrender","segmentation","点云","目标检测","遥感","3d","gan","geometry","jittorvis","jnerf","mesh","mlu","pytorch","分割","可微渲染","寒武纪","统信","飞腾"],"topics":[{"id":14,"title":"关于分享类别","fancy_title":"关于分享类别","slug":"topic","posts_count":2,"reply_count":0,"highest_post_number":2,"image_url":null,"created_at":"2021-09-29T14:43:29.597Z","last_posted_at":"2021-10-05T07:36:27.628Z","bumped":true,"bumped_at":"2021-09-29T14:52:21.466Z","archetype":"regular","unseen":false,"pinned":true,"unpinned":null,"excerpt":"来分享你或者你发现的前沿研究、模型库、计图工具吧。","visible":true,"closed":true,"archived":false,"bookmarked":null,"liked":null,"tags":[],"tags_descriptions":{},"views":678,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":905,"title":"pytorch 代码转 jittor 使用心得","fancy_title":"pytorch 代码转 jittor 使用心得","slug":"topic","posts_count":2,"reply_count":0,"highest_post_number":2,"image_url":null,"created_at":"2024-12-31T09:05:33.415Z","last_posted_at":"2025-10-14T14:20:10.795Z","bumped":true,"bumped_at":"2025-10-14T14:20:10.795Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"前言\n为了实现大作业中复现 jittor 代码的任务，这几天花了将近二十个小时在 jittor 环境的配置上，颇有一番折磨心得体会，在这里记录一下。 \n需要注意的是，jittor 与 torch 有一些关键的区别： \n\njittor 的基…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["模型库"],"tags_descriptions":{},"views":1799,"like_count":3,"has_summary":false,"last_poster_username":"moonClll","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":null,"description":"Original Poster","user_id":463,"primary_group_id":null,"flair_group_id":null},{"extras":"latest","description":"Most Recent Poster","user_id":516,"primary_group_id":null,"flair_group_id":null}]},{"id":405,"title":"新增四种新模型、支持NeRF转mesh，神经辐射场算法库JNeRF再升级","fancy_title":"新增四种新模型、支持NeRF转mesh，神经辐射场算法库JNeRF再升级","slug":"topic","posts_count":2,"reply_count":0,"highest_post_number":2,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/8507d578293045690d0fda26d69d5fc7292df1e5_2_1024x319.png","created_at":"2022-10-11T05:03:52.921Z","last_posted_at":"2022-10-31T10:55:09.238Z","bumped":true,"bumped_at":"2022-10-31T10:55:09.238Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"10月11日，计图(Jittor)团队开发的Jittor神经辐射场算法库JNeRF再次升级。这次JNeRF的改进，包括以下两个方面： \n\n\n新增了4个模型：Plenoxel(CVPR 2022 oral)、NeuS(NeurIPS 2021…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":[],"tags_descriptions":{},"views":7638,"like_count":0,"has_summary":false,"last_poster_username":"cuijianzhu","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":null,"description":"Original Poster","user_id":1,"primary_group_id":null,"flair_group_id":null},{"extras":"latest","description":"Most Recent Poster","user_id":179,"primary_group_id":null,"flair_group_id":null}]},{"id":412,"title":"JSparse：Jittor稀疏卷积计算库开源啦！","fancy_title":"JSparse：Jittor稀疏卷积计算库开源啦！","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/original/1X/5b9bb92f61cdb1874e4861df28d5762ac1373f45.jpeg","created_at":"2022-10-22T03:37:30.888Z","last_posted_at":"2022-10-22T03:37:30.999Z","bumped":true,"bumped_at":"2022-10-22T03:37:30.999Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"10月20日，计图团队发布了稀疏卷积计算库，可支持稀疏卷积，稀疏矩阵乘法等，该类能够支持3D物体分类，分割，识别和重建等任务。 \n计图(Jittor)是清华大学计算机系图形学实验室于2020年3月20日发布并开源的深度学习框架。计图团队已经…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":[],"tags_descriptions":{},"views":2298,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":132,"title":"计图开源：基于层次结构的隐式几何表示和建模","fancy_title":"计图开源：基于层次结构的隐式几何表示和建模","slug":"topic","posts_count":2,"reply_count":0,"highest_post_number":2,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/13c640ae209fbaf26538da51e7322bca3a78cebd_2_1024x430.webp","created_at":"2021-12-17T07:12:07.196Z","last_posted_at":"2022-10-20T13:40:14.670Z","bumped":true,"bumped_at":"2022-10-20T13:40:14.670Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"中科院计算所高林团队和腾讯游戏数字内容技术中心合作，在NeurIPS 2021上发表了一项研究工作OctField[1]，该工作提出了一种可微分递归编码的八叉树隐式场方法来表示和建模几何形状，达到了高精度的几何建模效果，并开源基于计图的实现…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["3d","geometry"],"tags_descriptions":{},"views":2600,"like_count":0,"has_summary":false,"last_poster_username":"我是一个jittor用户","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":null,"description":"Original Poster","user_id":1,"primary_group_id":null,"flair_group_id":null},{"extras":"latest","description":"Most Recent Poster","user_id":150,"primary_group_id":null,"flair_group_id":null}]},{"id":329,"title":"常见 Torch 与 Jittor 接口的对应关系","fancy_title":"常见 Torch 与 Jittor 接口的对应关系","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":null,"created_at":"2022-08-14T15:15:50.491Z","last_posted_at":"2022-08-14T15:15:50.608Z","bumped":true,"bumped_at":"2022-08-14T15:17:01.335Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"由于设计理念和底层逻辑的区别，Jittor 中有部分接口与 Torch 是不一样的。在这里整理如下，以供参考。 \n\n\n\n\nTorch\nJittor\n解释\n\n\n\n\ntorch.autograd.variable()\n无\njt.Var 支持求导…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["pytorch"],"tags_descriptions":{},"views":2415,"like_count":2,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":275,"title":"Jittor开源: JNeRF带你5秒训好NeRF","fancy_title":"Jittor开源: JNeRF带你5秒训好NeRF","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/072d27da6c841f0f44ef761ea4b0fd23c93e2f4b_2_1024x580.jpeg","created_at":"2022-06-01T13:35:32.968Z","last_posted_at":"2022-06-01T13:35:33.155Z","bumped":true,"bumped_at":"2022-06-01T13:35:33.155Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"计图(Jittor)框架的NeRF模型库JNeRF正式开源了！ \n通过JNeRF可以5秒训练好NeRF模型(见图1)。 \nJittor成为首个支持Instant-NGP的深度学习框架。 \n  \n图1 5秒训练NeRF效果展示 \n\nPart1…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["jnerf"],"tags_descriptions":{},"views":2688,"like_count":0,"has_summary":false,"last_poster_username":"system","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":-1,"primary_group_id":null,"flair_group_id":null}]},{"id":134,"title":"清华大学将召开首届计图开发者交流会","fancy_title":"清华大学将召开首届计图开发者交流会","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/e40191f84fb58941d66a0711b3efc8e0552e12e5_2_1024x582.jpeg","created_at":"2021-12-19T07:04:39.657Z","last_posted_at":"2021-12-19T07:04:39.918Z","bumped":true,"bumped_at":"2021-12-19T07:04:39.918Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"2020年3月20日，计图 (Jittor) 深度学习框架开源。为促进人工智能技术的发展和开源深度学习生态的构建，Jittor团队已举办五期深度学习研讨会，并成功举办了第一届计图挑战赛。 \n为了进一步推进Jittor深度学习框架的开源生态构…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":[],"tags_descriptions":{},"views":1438,"like_count":1,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":110,"title":"支持SAR图像、新增可视化接口，遥感目标检测库JDet再升级","fancy_title":"支持SAR图像、新增可视化接口，遥感目标检测库JDet再升级","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/840ce3e4ad9d782471924a3038eb1e9c9c742813_2_1024x518.webp","created_at":"2021-11-17T08:33:45.138Z","last_posted_at":"2021-11-17T08:33:45.352Z","bumped":true,"bumped_at":"2021-11-17T08:33:45.352Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"支持SAR图像、新增可视化接口，遥感目标检测库JDet再升级\n11月17日，清华大学计图 (Jittor) 团队开发的Jittor遥感图像目标检测算法库JDet再次升级。这次JDet的改进，包括以下四个方面： \n\n\n新增了4个模型：Ori…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":[],"tags_descriptions":{},"views":2100,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":104,"title":"开发针对jittor原算子的asic？","fancy_title":"开发针对jittor原算子的asic？","slug":"topic","posts_count":3,"reply_count":0,"highest_post_number":3,"image_url":null,"created_at":"2021-11-07T02:09:46.627Z","last_posted_at":"2021-11-09T23:56:20.037Z","bumped":true,"bumped_at":"2021-11-09T23:56:20.037Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"如果想开发jittor的定制的asic，不知道是否有可行性，以及性能上的飞跃，相对于cpu、gpu等。 \n比如把元算子用verilog实现，计算资源，内存，调度等也硬件定制化。。。","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":[],"tags_descriptions":{},"views":1238,"like_count":3,"has_summary":false,"last_poster_username":"yyz","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest","description":"Original Poster, Most Recent Poster","user_id":25,"primary_group_id":null,"flair_group_id":null},{"extras":null,"description":"Frequent Poster","user_id":2,"primary_group_id":null,"flair_group_id":null}]},{"id":32,"title":"[模型库] Jittor开源点云模型库，多种主流模型训练比Pytorch提升一倍以上","fancy_title":"[模型库] Jittor开源点云模型库，多种主流模型训练比Pytorch提升一倍以上","slug":"topic","posts_count":3,"reply_count":1,"highest_post_number":3,"image_url":"https://discuss.jittor.cn/uploads/default/original/1X/7c1524024607d78740ac79cb76dd71edac23fe3c.webp","created_at":"2021-10-04T08:58:14.653Z","last_posted_at":"2021-10-18T03:37:40.400Z","bumped":true,"bumped_at":"2021-10-18T03:37:40.400Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"2020年10月10日，第二届清华大学计图深度学习研讨会成功举办，研讨会上，计图还发布了检测与分割模型库、点云模型库和可微渲染功能。 \n点云数据的获取与处理是图形学和三维视觉中的重要问题，在测绘、自动驾驶等方面有着广泛的应用。由于其巨大的应…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["模型库","official","点云"],"tags_descriptions":{},"views":2680,"like_count":0,"has_summary":false,"last_poster_username":"MenghaoGuo","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":null,"description":"Original Poster","user_id":1,"primary_group_id":null,"flair_group_id":null},{"extras":null,"description":"Frequent Poster","user_id":9,"primary_group_id":null,"flair_group_id":null},{"extras":"latest","description":"Most Recent Poster","user_id":8,"primary_group_id":null,"flair_group_id":null}]},{"id":57,"title":"计图开源：纹理网格模型生成方法TM-NET","fancy_title":"计图开源：纹理网格模型生成方法TM-NET","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/c6b8fb37ca3deac3b2faf24db44c921f64f2018f_2_1024x451.png","created_at":"2021-10-04T11:04:39.582Z","last_posted_at":"2021-10-04T11:04:40.001Z","bumped":true,"bumped_at":"2021-10-05T06:52:44.000Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"计图开源：纹理网格模型生成方法TM-NET\n近期中科院计算所高林团队和SFU的张皓教授、英国卡迪夫大学来煜坤教授合作，在计算机图形学顶会ACM SIGGRAPH Asia 2021上发表纹理网格模型生成方法TM-NET[1]，并在计图框架…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["research"],"tags_descriptions":{},"views":1524,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":56,"title":"计图开源：邵岭教授团队提出息肉分割网络PraNet，计图框架下推理速度大幅提升","fancy_title":"计图开源：邵岭教授团队提出息肉分割网络PraNet，计图框架下推理速度大幅提升","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/f45de18a57a8ab6eec7088475f675294722e4c62_2_1024x514.png","created_at":"2021-10-04T11:03:46.548Z","last_posted_at":"2021-10-04T11:03:46.895Z","bumped":true,"bumped_at":"2021-10-04T11:03:46.895Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"邵岭教授团队提出息肉分割网络PraNet，计图框架下推理速度大幅提升\n计图 (Jittor) 是清华大学自主研发，于2020年3月开源的深度学习框架。Jittor框架因其底层使用了元算子和统一计算图，从而大大提升了深度学习框架的性能和灵活…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["research"],"tags_descriptions":{},"views":1450,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":55,"title":"计图遥感库再升级，助力“中科星图杯”高分遥感图像解译大赛”分割”赛道","fancy_title":"计图遥感库再升级，助力“中科星图杯”高分遥感图像解译大赛”分割”赛道","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/b93bc7aa307035141d2aca5b10ca5d5117b30da8_2_1024x576.png","created_at":"2021-10-04T11:02:53.649Z","last_posted_at":"2021-10-04T11:02:54.013Z","bumped":true,"bumped_at":"2021-10-04T11:02:54.013Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"计图(Jittor)是清华大学自主研发的深度学习框架，是中国科学院空天信息创新研究院主办的第五届“中科星图杯”国际高分遥感图像解译大赛官方指定深度学习框架之一，Jittor团队针对大赛所使用的数据集和提交方式提供了相关支持和工具。 \n大赛地…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["jdet","遥感"],"tags_descriptions":{},"views":1311,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":54,"title":"计图遥感库JDet助力“中科星图杯”高分遥感图像解译大赛","fancy_title":"计图遥感库JDet助力“中科星图杯”高分遥感图像解译大赛","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/db634e651d19fc9b6b36534fe610912510868de8_2_1024x512.png","created_at":"2021-10-04T11:00:55.863Z","last_posted_at":"2021-10-04T11:00:56.103Z","bumped":true,"bumped_at":"2021-10-04T11:00:56.103Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"计图(Jittor)是清华大学自主研发的深度学习框架，是中国科学院空天信息创新研究院主办的第五届“中科星图杯”国际高分遥感图像解译大赛官方指定深度学习框架之一，Jittor团队针对大赛所使用的数据集和提交方式提供了相关支持和工具。 \n大赛地…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["jdet","遥感"],"tags_descriptions":{},"views":1769,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":53,"title":"JMedSeg: Jittor医学图像智能分割模型库开源了！","fancy_title":"JMedSeg: Jittor医学图像智能分割模型库开源了！","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/758caf6cf553645af2f21ac541ff9e7782a38b46_2_1024x683.png","created_at":"2021-10-04T10:58:49.486Z","last_posted_at":"2021-10-04T10:58:49.773Z","bumped":true,"bumped_at":"2021-10-04T10:58:49.773Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"在北京协和医院核医学科主任霍力教授、清华长庚医院核医学科主任何作祥教授的共同指导下，清华大学计算机系陶霖密老师及其《模式识别》课程的三十多位同学的倾心之作，基于深度学习框架计图(Jittor)的医学图像智能分割模型库JMedSeg发布了！ \n…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["模型库","segmentation","分割"],"tags_descriptions":{},"views":2068,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":52,"title":"计图开源：三维几何模型的可微变形分析与编辑方法","fancy_title":"计图开源：三维几何模型的可微变形分析与编辑方法","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/original/1X/7d969043081bc49ca0e2d9677ef9af38878991b3.png","created_at":"2021-10-04T10:57:51.408Z","last_posted_at":"2021-10-04T10:57:51.716Z","bumped":true,"bumped_at":"2021-10-04T10:57:51.716Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"近期，中科院计算所高林团队于IEEE TPAMI发表了一项重要的研究成果[1]，提出了一种新颖的三维模型局部变形基底分析方法和可微几何模型编辑建模方法，并已开源了基于计图实现的代码，网络模型的训练速度较Tensorflow提升53%。 \nP…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["research"],"tags_descriptions":{},"views":1049,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":51,"title":"计图开源：基于标准化生成流的人体运动风格迁移方法","fancy_title":"计图开源：基于标准化生成流的人体运动风格迁移方法","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/original/1X/7db8eb01451ba4fa2d148c0fc60069f6b33aeb81.png","created_at":"2021-10-04T10:57:18.557Z","last_posted_at":"2021-10-04T10:57:18.852Z","bumped":true,"bumped_at":"2021-10-04T10:57:18.852Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"近期，清华大学温玉辉博士后、刘永进教授、中科院计算所副研究员高林、香港城市大学傅红波教授等合作，在CVPR2021上发表论文，提出了一种基于标准化生成流（Glow）的自回归运动风格迁移方法，并在GitHub上开源了Jittor代码。 \nPa…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["research"],"tags_descriptions":{},"views":1333,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":50,"title":"支持遥感图像了，Jittor开源物体检测算法库JDet","fancy_title":"支持遥感图像了，Jittor开源物体检测算法库JDet","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/50e525b54aaba5863af2e1485d8b9eb4747654f5_2_1024x515.png","created_at":"2021-10-04T09:21:47.337Z","last_posted_at":"2021-10-04T09:21:47.673Z","bumped":true,"bumped_at":"2021-10-04T09:21:47.673Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"8月4日，清华大学计图(Jittor)团队发布了遥感图像物体检测算法库JDet，支持S2ANet，Gliding，RetinaNet,Faster R-CNN等主流的遥感图像物体检测模型。希望借助Jittor的性能优势和能较好地支持国产操作…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["模型库","official","jdet","detection","目标检测"],"tags_descriptions":{},"views":1821,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":49,"title":"计图开源：基于梯度编辑的鲁棒性人脸识别方法SFace","fancy_title":"计图开源：基于梯度编辑的鲁棒性人脸识别方法SFace","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/e85d9fa34f2bcd27903ad7ca9ceb44180642a599_2_1024x334.png","created_at":"2021-10-04T09:20:44.062Z","last_posted_at":"2021-10-04T09:20:44.437Z","bumped":true,"bumped_at":"2021-10-04T09:20:44.437Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"人脸识别是一项经典的人工智能应用。得益于海量的训练数据和深度学习的技术发展，人脸识别取得了突破性的性能提升，成为一项广泛应用的生物识别技术。 \n但人脸识别仍然存在诸多挑战：高质量的训练数据是高性能识别模型的基础，然而目前学术界可用的数据集存…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["research"],"tags_descriptions":{},"views":1397,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":48,"title":"计图开源：基于层次布局感知图卷积网络的图像美学评估方法HLA-GCN","fancy_title":"计图开源：基于层次布局感知图卷积网络的图像美学评估方法HLA-GCN","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/994dcb94246f3986922b92b197566398b5707ffe_2_1024x342.png","created_at":"2021-10-04T09:19:28.200Z","last_posted_at":"2021-10-04T09:19:28.695Z","bumped":true,"bumped_at":"2021-10-04T09:19:28.695Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"随着人工智能的发展，计算机在识物辨人等方面的能力与人类水平的差距逐渐缩小，而如何使计算机具备审美能力则仍然是一个很有挑战性的难题。图像美学质量评估（Image Aesthetics Assessment,IAA）就是希望通过计算机来模拟人类…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["research"],"tags_descriptions":{},"views":1659,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":47,"title":"计图开源： 程明明团队提出类别激活图方法LayerCAM","fancy_title":"计图开源： 程明明团队提出类别激活图方法LayerCAM","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/80929c3cea1bf3777f94e4bb88d822aec17a9fba_2_1024x561.png","created_at":"2021-10-04T09:18:54.658Z","last_posted_at":"2021-10-04T09:18:55.089Z","bumped":true,"bumped_at":"2021-10-04T09:18:55.089Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"类别激活图是从分类网络中生成的类别响应图，它可以在图像类别标签级标注信息的基础上粗略定位到图像中具有判别性的物体区域。近期，南开大学媒体计算实验室程明明教授团队提出了一种层次化类别激活图的生成方法，称为LayerCAM[1]。已有类别激活图…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["research"],"tags_descriptions":{},"views":2889,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":46,"title":"计图团队联合发布可视化工具JittorVis，帮助开发者理解深度神经网络模型","fancy_title":"计图团队联合发布可视化工具JittorVis，帮助开发者理解深度神经网络模型","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/3af06682eddb9bebaed8a664b5b3a936ef16e2fc_2_1024x510.jpeg","created_at":"2021-10-04T09:18:18.306Z","last_posted_at":"2021-10-04T09:18:18.515Z","bumped":true,"bumped_at":"2021-10-04T09:18:18.515Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"6月30日，清华大学计图（Jittor）团队与刘世霞教授课题组合作开发的基于Jittor的可视化工具JittorVis正式发布。该工具用于帮助理解基于Jittor构建的深度神经网络模型的计算图结构和数据流信息。 \nPart 1 \n计算图可视…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["jittorvis"],"tags_descriptions":{},"views":1592,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":45,"title":"计图发布Jrender 2.0，可微渲染再升级，NeRF应用两倍提速！","fancy_title":"计图发布Jrender 2.0，可微渲染再升级，NeRF应用两倍提速！","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/61c46c3fce52a5eced90c8f3cba94ddf02912b96_2_1024x318.png","created_at":"2021-10-04T09:17:31.698Z","last_posted_at":"2021-10-04T09:17:32.228Z","bumped":true,"bumped_at":"2021-10-04T09:17:32.228Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"6月28日，清华大学计图(Jittor)团队发布了可微渲染库Jrender 2.0新版本，正式支持可微体渲染！近期学术界已有大量工作基于可微体渲染完成，其中备受关注的NeRF[1]应用，在Jrender的加速下训练、测试速度增幅可达2倍！ \n…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["模型库","jrender","official"],"tags_descriptions":{},"views":2467,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":44,"title":"计图支持国产统信操作系统和飞腾CPU芯片，实现GPU加速的训练和推理","fancy_title":"计图支持国产统信操作系统和飞腾CPU芯片，实现GPU加速的训练和推理","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/f04ea9baa36548df21cd650171dffffd615cf814_2_1024x576.png","created_at":"2021-10-04T09:16:17.035Z","last_posted_at":"2021-10-04T09:16:17.350Z","bumped":true,"bumped_at":"2021-10-04T09:16:17.350Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"计图(Jittor)是由清华大学发布的首个动态编译深度学习框架，在5月8日兼容寒武纪芯片之后，计图目前又在国内自主研发操作系统统信UOS+国产CPU飞腾FT2000的架构下，成功实现了CPU和GPU的训练推理。 \n  \n图1 使用统信操作系…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["official","飞腾","统信"],"tags_descriptions":{},"views":2119,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":43,"title":"计图开源：计图团队首创三角网格面片上的卷积神经网络、图像上的网络架构可以做三维模型的深度学习了！","fancy_title":"计图开源：计图团队首创三角网格面片上的卷积神经网络、图像上的网络架构可以做三维模型的深度学习了！","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/81261e66bf186dbb508f47e81c2aa3089fef8c09_2_1024x305.png","created_at":"2021-10-04T09:15:05.021Z","last_posted_at":"2021-10-04T09:15:05.486Z","bumped":true,"bumped_at":"2021-10-04T09:15:05.486Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"近日，清华大学计图(Jittor)团队提出了一种针对三角网格的卷积神经网络，在两个网格分类数据集上首次取得100%正确率，在其他多个几何学习任务中，性能显著超过现有方法。 \n尤为重要的是，这种基于细分表示的网格卷积神经网络的提出，使得VGG…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["research","mesh"],"tags_descriptions":{},"views":1484,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":42,"title":"计图开源：隐蔽目标检测新任务在计图框架下推理性能大幅提升","fancy_title":"计图开源：隐蔽目标检测新任务在计图框架下推理性能大幅提升","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/9aed2fa8f7eddad1fd51d4ba3918821b9245208e_2_1024x345.png","created_at":"2021-10-04T09:13:59.326Z","last_posted_at":"2021-10-04T09:13:59.929Z","bumped":true,"bumped_at":"2021-10-04T09:13:59.929Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"计图(Jittor)深度学习框架于2020年3月由清华大学自主研发并发布，在诸多应用领域取得了优于国际主流框架的性能。 \n近期，阿联酋起源人工智能研究院（IIAI）邵岭教授团队和南开大学程明明教授团队联合构建了隐蔽目标检测（Conceale…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["research","detection","目标检测"],"tags_descriptions":{},"views":1136,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":41,"title":"计图开源：程明明教授团队提出深度霍夫变换，Jittor框架下语义线检测性能大幅提升","fancy_title":"计图开源：程明明教授团队提出深度霍夫变换，Jittor框架下语义线检测性能大幅提升","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/b2e0410ca0ee3b08b2411d4c44e8c200823fa530_2_1024x350.png","created_at":"2021-10-04T09:13:03.550Z","last_posted_at":"2021-10-04T09:13:03.828Z","bumped":true,"bumped_at":"2021-10-04T09:13:03.828Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"计图(Jittor)是清华大学自主研发的深度学习框架，自2020年3月发布以来，因为元算子融合和统一计算图的两大创新，在很多应用领域，取得了优于国际主流框架的性能。 \n近期，南开大学程明明教授团队开源了基于Jittor的语义线检测算法，称为…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["research"],"tags_descriptions":{},"views":1724,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":40,"title":"计图开源：无监督三维模型对称性分析方法PRS-NET","fancy_title":"计图开源：无监督三维模型对称性分析方法PRS-NET","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/optimized/1X/2b4fb49890b2b8a5b1b2e7309940fd9006ee11ef_2_1024x146.png","created_at":"2021-10-04T09:12:11.092Z","last_posted_at":"2021-10-04T09:12:11.263Z","bumped":true,"bumped_at":"2021-10-04T09:12:11.263Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"对称性是三维模型的重要属性之一，对三维模型进行对称性分析，可以用于几何模型识别、分析和建模等任务。中科院计算所在计算机图形学顶级期刊IEEE TVCG上发文，提出了一种无监督的三维模型对称性分析方法PRS-NET，该方法利用了对称模型本身具…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["research"],"tags_descriptions":{},"views":1388,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]},{"id":39,"title":"计图支持寒武纪！动态图性能较PyTorch提升270倍","fancy_title":"计图支持寒武纪！动态图性能较PyTorch提升270倍","slug":"topic","posts_count":1,"reply_count":0,"highest_post_number":1,"image_url":"https://discuss.jittor.cn/uploads/default/original/1X/3408af111c2abfb4e2f489df6287605b22203da6.png","created_at":"2021-10-04T09:11:32.062Z","last_posted_at":"2021-10-04T09:11:32.275Z","bumped":true,"bumped_at":"2021-10-04T09:11:32.275Z","archetype":"regular","unseen":false,"pinned":false,"unpinned":null,"excerpt":"近日，计图(Jittor)团队成功完成寒武纪芯片MLU270上的移植，支持推理和训练，并复现了ResNet、Alexnet、VGG等骨干网络，其中动态图推理速度相较PyTorch平均提升了276.69倍，推理精度也得到了显著的提升。 \n  \n…","visible":true,"closed":false,"archived":false,"bookmarked":null,"liked":null,"tags":["寒武纪","mlu"],"tags_descriptions":{},"views":1502,"like_count":0,"has_summary":false,"last_poster_username":"lzhengning","category_id":6,"pinned_globally":false,"featured_link":null,"has_accepted_answer":false,"can_have_answer":false,"posters":[{"extras":"latest single","description":"Original Poster, Most Recent Poster","user_id":1,"primary_group_id":null,"flair_group_id":null}]}]}}