From 57d80b4d778a65edf005cb2d7cb784dfdcfb0fed Mon Sep 17 00:00:00 2001 From: Tianliang Zhang Date: Wed, 12 Jun 2024 23:05:04 +1000 Subject: [PATCH] Site updated: 2024-06-12 23:05:03 --- .../14/Generative-Adversarial-Nets/index.html | 2 +- 2018/08/14/How-to-use-hexo/index.html | 2 +- 2018/08/14/Keras-Tutorial/index.html | 2 +- .../Pedestrian-Detection-Sources/index.html | 2 +- .../index.html | 2 +- .../09/05/MLIA-Logistic-Regression/index.html | 2 +- 2018/09/09/Conda-Tutorials/index.html | 2 +- 2018/10/10/docker/index.html | 2 +- 2019/01/10/Focal-Loss/index.html | 2 +- 2019/04/11/Pillow-Tutorial/index.html | 2 +- .../index.html | 2 +- 2019/05/20/Adaptive-NMS/index.html | 2 +- 2019/05/23/CVPR19-HRNet/index.html | 2 +- 2019/05/29/CVPR2019-CSP/index.html | 2 +- 2019/06/20/Book-BianLiang/index.html | 2 +- 2019/06/23/Pandas-Tutorial/index.html | 2 +- 2019/07/10/l2normalization/index.html | 2 +- 2019/07/13/Beancount-01/index.html | 2 +- 2021/11/24/Deformable-DETR/index.html | 2 +- 2021/12/27/Efficient-DETR/index.html | 2 +- 2022/05/19/SOTR/index.html | 2 +- 2022/05/20/K-Net/index.html | 2 +- about/index.html | 2 +- archives/2018/08/index.html | 2 +- archives/2018/09/index.html | 2 +- archives/2018/10/index.html | 2 +- archives/2018/index.html | 2 +- archives/2019/01/index.html | 2 +- archives/2019/04/index.html | 2 +- archives/2019/05/index.html | 2 +- archives/2019/06/index.html | 2 +- archives/2019/07/index.html | 2 +- archives/2019/index.html | 2 +- archives/2021/11/index.html | 2 +- archives/2021/12/index.html | 2 +- archives/2021/index.html | 2 +- archives/2022/05/index.html | 2 +- archives/2022/index.html | 2 +- archives/index.html | 2 +- archives/page/2/index.html | 2 +- archives/page/3/index.html | 2 +- categories/Book/index.html | 2 +- categories/Deep-Learning/index.html | 2 +- categories/Machine-Learning/index.html | 2 +- categories/Object-Detection/index.html | 2 +- categories/Pedestrian-Detection/index.html | 2 +- categories/Python/index.html | 2 +- categories/Tools/index.html | 2 +- categories/index.html | 2 +- css/main.css | 2 +- index.html | 2 +- page/2/index.html | 2 +- page/3/index.html | 2 +- search.xml | 402 +++++++++--------- tags/Beancount/index.html | 2 +- tags/Book/index.html | 2 +- tags/Deep-Learning/index.html | 2 +- tags/GAN/index.html | 2 +- tags/Instance-Segmentation/index.html | 2 +- tags/Keras/index.html | 2 +- tags/Object-Detection/index.html | 2 +- tags/Pandas/index.html | 2 +- tags/Pedestrian-Detection/index.html | 2 +- tags/Pose-Estimation/index.html | 2 +- tags/Python/index.html | 2 +- tags/Tools/index.html | 2 +- tags/Transformer/index.html | 2 +- tags/docker/index.html | 2 +- tags/index.html | 2 +- 69 files changed, 269 insertions(+), 269 deletions(-) diff --git a/2018/08/14/Generative-Adversarial-Nets/index.html b/2018/08/14/Generative-Adversarial-Nets/index.html index 6c6f6dd..f8b372f 100644 --- a/2018/08/14/Generative-Adversarial-Nets/index.html +++ b/2018/08/14/Generative-Adversarial-Nets/index.html @@ -429,7 +429,7 @@

Generative Adversarial - + diff --git a/2018/08/14/How-to-use-hexo/index.html b/2018/08/14/How-to-use-hexo/index.html index 5429da2..dfcad24 100644 --- a/2018/08/14/How-to-use-hexo/index.html +++ b/2018/08/14/How-to-use-hexo/index.html @@ -469,7 +469,7 @@

Hexo 资源

- + diff --git a/2018/08/14/Keras-Tutorial/index.html b/2018/08/14/Keras-Tutorial/index.html index ac8ced1..1d18a90 100644 --- a/2018/08/14/Keras-Tutorial/index.html +++ b/2018/08/14/Keras-Tutorial/index.html @@ -438,7 +438,7 @@

目录

- + diff --git a/2018/08/17/Pedestrian-Detection-Sources/index.html b/2018/08/17/Pedestrian-Detection-Sources/index.html index d3d0591..8c0a497 100644 --- a/2018/08/17/Pedestrian-Detection-Sources/index.html +++ b/2018/08/17/Pedestrian-Detection-Sources/index.html @@ -1140,7 +1140,7 @@

性能比较

- + diff --git a/2018/09/02/CornerNet-Detection-Objects-as-Paired-Keypoints/index.html b/2018/09/02/CornerNet-Detection-Objects-as-Paired-Keypoints/index.html index 18d13a5..2a42de1 100644 --- a/2018/09/02/CornerNet-Detection-Objects-as-Paired-Keypoints/index.html +++ b/2018/09/02/CornerNet-Detection-Objects-as-Paired-Keypoints/index.html @@ -419,7 +419,7 @@

- + diff --git a/2018/09/05/MLIA-Logistic-Regression/index.html b/2018/09/05/MLIA-Logistic-Regression/index.html index 2e78372..f90a07c 100644 --- a/2018/09/05/MLIA-Logistic-Regression/index.html +++ b/2018/09/05/MLIA-Logistic-Regression/index.html @@ -506,7 +506,7 @@

可视化权重(weights)的变化

- + diff --git a/2018/09/09/Conda-Tutorials/index.html b/2018/09/09/Conda-Tutorials/index.html index fcf8b79..3acba50 100644 --- a/2018/09/09/Conda-Tutorials/index.html +++ b/2018/09/09/Conda-Tutorials/index.html @@ -430,7 +430,7 @@

Miniconda　镜像

- + diff --git a/2018/10/10/docker/index.html b/2018/10/10/docker/index.html index d497af3..619d8e0 100644 --- a/2018/10/10/docker/index.html +++ b/2018/10/10/docker/index.html @@ -432,7 +432,7 @@

5. 学习资源

- + diff --git a/2019/01/10/Focal-Loss/index.html b/2019/01/10/Focal-Loss/index.html index e5c3101..4d7d9ec 100644 --- a/2019/01/10/Focal-Loss/index.html +++ b/2019/01/10/Focal-Loss/index.html @@ -438,7 +438,7 @@

3.4 Class Imbalance and - + diff --git a/2019/04/11/Pillow-Tutorial/index.html b/2019/04/11/Pillow-Tutorial/index.html index fe1559f..6541024 100644 --- a/2019/04/11/Pillow-Tutorial/index.html +++ b/2019/04/11/Pillow-Tutorial/index.html @@ -642,7 +642,7 @@

使用草稿（draft）模式读取

- + diff --git a/2019/05/17/Squeeze-and-Excitation-Networks/index.html b/2019/05/17/Squeeze-and-Excitation-Networks/index.html index 7f006df..88cee88 100644 --- a/2019/05/17/Squeeze-and-Excitation-Networks/index.html +++ b/2019/05/17/Squeeze-and-Excitation-Networks/index.html @@ -435,7 +435,7 @@

Pytorch实现SE模块

- + diff --git a/2019/05/20/Adaptive-NMS/index.html b/2019/05/20/Adaptive-NMS/index.html index 7333a32..ac9f2b2 100644 --- a/2019/05/20/Adaptive-NMS/index.html +++ b/2019/05/20/Adaptive-NMS/index.html @@ -447,7 +447,7 @@

Experiments

- + diff --git a/2019/05/23/CVPR19-HRNet/index.html b/2019/05/23/CVPR19-HRNet/index.html index 11df08a..843da43 100644 --- a/2019/05/23/CVPR19-HRNet/index.html +++ b/2019/05/23/CVPR19-HRNet/index.html @@ -458,7 +458,7 @@

HighResolutionModule

- + diff --git a/2019/05/29/CVPR2019-CSP/index.html b/2019/05/29/CVPR2019-CSP/index.html index 97d5b50..86cf423 100644 --- a/2019/05/29/CVPR2019-CSP/index.html +++ b/2019/05/29/CVPR2019-CSP/index.html @@ -476,7 +476,7 @@

Loss

- + diff --git a/2019/06/20/Book-BianLiang/index.html b/2019/06/20/Book-BianLiang/index.html index 3f905df..e620cf0 100644 --- a/2019/06/20/Book-BianLiang/index.html +++ b/2019/06/20/Book-BianLiang/index.html @@ -396,7 +396,7 @@

第五章　阿那亚和范家小学 - + diff --git a/2019/06/23/Pandas-Tutorial/index.html b/2019/06/23/Pandas-Tutorial/index.html index bce2300..ee9fd0f 100644 --- a/2019/06/23/Pandas-Tutorial/index.html +++ b/2019/06/23/Pandas-Tutorial/index.html @@ -466,7 +466,7 @@

直方图（Histograms）

- + diff --git a/2019/07/10/l2normalization/index.html b/2019/07/10/l2normalization/index.html index 207957f..06d9c9f 100644 --- a/2019/07/10/l2normalization/index.html +++ b/2019/07/10/l2normalization/index.html @@ -417,7 +417,7 @@

Pytorch Code

- + diff --git a/2019/07/13/Beancount-01/index.html b/2019/07/13/Beancount-01/index.html index 8c9775b..e3bf594 100644 --- a/2019/07/13/Beancount-01/index.html +++ b/2019/07/13/Beancount-01/index.html @@ -561,7 +561,7 @@

其他介绍文章

- + diff --git a/2021/11/24/Deformable-DETR/index.html b/2021/11/24/Deformable-DETR/index.html index 0b9fd2f..508011a 100644 --- a/2021/11/24/Deformable-DETR/index.html +++ b/2021/11/24/Deformable-DETR/index.html @@ -572,7 +572,7 @@

论文中的符号说明

- + diff --git a/2021/12/27/Efficient-DETR/index.html b/2021/12/27/Efficient-DETR/index.html index 17e6ea7..5629edc 100644 --- a/2021/12/27/Efficient-DETR/index.html +++ b/2021/12/27/Efficient-DETR/index.html @@ -573,7 +573,7 @@

实验部分

- + diff --git a/2022/05/19/SOTR/index.html b/2022/05/19/SOTR/index.html index ce2755d..2685aad 100644 --- a/2022/05/19/SOTR/index.html +++ b/2022/05/19/SOTR/index.html @@ -511,7 +511,7 @@

实验

- + diff --git a/2022/05/20/K-Net/index.html b/2022/05/20/K-Net/index.html index 9705b91..53a7b81 100644 --- a/2022/05/20/K-Net/index.html +++ b/2022/05/20/K-Net/index.html @@ -512,7 +512,7 @@

总结

- + diff --git a/about/index.html b/about/index.html index 4c53b54..4fa9a8c 100644 --- a/about/index.html +++ b/about/index.html @@ -293,7 +293,7 @@

Tianliang Zhang - + diff --git a/archives/2018/08/index.html b/archives/2018/08/index.html index 65e38b2..74442dc 100644 --- a/archives/2018/08/index.html +++ b/archives/2018/08/index.html @@ -349,7 +349,7 @@

Tianliang

- + diff --git a/archives/2018/09/index.html b/archives/2018/09/index.html index c1a86a8..ff85d7e 100644 --- a/archives/2018/09/index.html +++ b/archives/2018/09/index.html @@ -329,7 +329,7 @@

Tianliang

- + diff --git a/archives/2018/10/index.html b/archives/2018/10/index.html index d16ddcd..cddfb43 100644 --- a/archives/2018/10/index.html +++ b/archives/2018/10/index.html @@ -289,7 +289,7 @@

Tianliang

- + diff --git a/archives/2018/index.html b/archives/2018/index.html index 5089370..7c90b41 100644 --- a/archives/2018/index.html +++ b/archives/2018/index.html @@ -429,7 +429,7 @@

Tianliang

- + diff --git a/archives/2019/01/index.html b/archives/2019/01/index.html index 0583326..2bf8a25 100644 --- a/archives/2019/01/index.html +++ b/archives/2019/01/index.html @@ -289,7 +289,7 @@

Tianliang

- + diff --git a/archives/2019/04/index.html b/archives/2019/04/index.html index 2eae135..6ab80e2 100644 --- a/archives/2019/04/index.html +++ b/archives/2019/04/index.html @@ -289,7 +289,7 @@

Tianliang

- + diff --git a/archives/2019/05/index.html b/archives/2019/05/index.html index a9a2010..4b8eb7f 100644 --- a/archives/2019/05/index.html +++ b/archives/2019/05/index.html @@ -349,7 +349,7 @@

Tianliang

- + diff --git a/archives/2019/06/index.html b/archives/2019/06/index.html index 499607b..eee34bb 100644 --- a/archives/2019/06/index.html +++ b/archives/2019/06/index.html @@ -309,7 +309,7 @@

Tianliang

- + diff --git a/archives/2019/07/index.html b/archives/2019/07/index.html index 751cf03..75a2628 100644 --- a/archives/2019/07/index.html +++ b/archives/2019/07/index.html @@ -309,7 +309,7 @@

Tianliang

- + diff --git a/archives/2019/index.html b/archives/2019/index.html index f97ce4b..2ca4aec 100644 --- a/archives/2019/index.html +++ b/archives/2019/index.html @@ -469,7 +469,7 @@

Tianliang

- + diff --git a/archives/2021/11/index.html b/archives/2021/11/index.html index 4133363..4bf1827 100644 --- a/archives/2021/11/index.html +++ b/archives/2021/11/index.html @@ -289,7 +289,7 @@

Tianliang

- + diff --git a/archives/2021/12/index.html b/archives/2021/12/index.html index 0efad0c..ede88be 100644 --- a/archives/2021/12/index.html +++ b/archives/2021/12/index.html @@ -289,7 +289,7 @@

Tianliang

- + diff --git a/archives/2021/index.html b/archives/2021/index.html index b89233e..9196520 100644 --- a/archives/2021/index.html +++ b/archives/2021/index.html @@ -309,7 +309,7 @@

Tianliang

- + diff --git a/archives/2022/05/index.html b/archives/2022/05/index.html index 5dec97e..8738309 100644 --- a/archives/2022/05/index.html +++ b/archives/2022/05/index.html @@ -309,7 +309,7 @@

Tianliang

- + diff --git a/archives/2022/index.html b/archives/2022/index.html index 08b0145..f4dd9a1 100644 --- a/archives/2022/index.html +++ b/archives/2022/index.html @@ -309,7 +309,7 @@

Tianliang

- + diff --git a/archives/index.html b/archives/index.html index 4b521df..1e1b1ed 100644 --- a/archives/index.html +++ b/archives/index.html @@ -478,7 +478,7 @@

Tianliang

- + diff --git a/archives/page/2/index.html b/archives/page/2/index.html index 595d154..60ea55b 100644 --- a/archives/page/2/index.html +++ b/archives/page/2/index.html @@ -475,7 +475,7 @@

Tianliang

- + diff --git a/archives/page/3/index.html b/archives/page/3/index.html index 2ff66c7..a1b1432 100644 --- a/archives/page/3/index.html +++ b/archives/page/3/index.html @@ -312,7 +312,7 @@

Tianliang

- + diff --git a/categories/Book/index.html b/categories/Book/index.html index 1c1ce06..d293920 100644 --- a/categories/Book/index.html +++ b/categories/Book/index.html @@ -290,7 +290,7 @@

Book - + diff --git a/categories/Deep-Learning/index.html b/categories/Deep-Learning/index.html index a4b8472..785878b 100644 --- a/categories/Deep-Learning/index.html +++ b/categories/Deep-Learning/index.html @@ -373,7 +373,7 @@

Deep Learning - + diff --git a/categories/Machine-Learning/index.html b/categories/Machine-Learning/index.html index ccc2ceb..8bdbc72 100644 --- a/categories/Machine-Learning/index.html +++ b/categories/Machine-Learning/index.html @@ -290,7 +290,7 @@

Machine Learning - + diff --git a/categories/Object-Detection/index.html b/categories/Object-Detection/index.html index 419e068..7e54465 100644 --- a/categories/Object-Detection/index.html +++ b/categories/Object-Detection/index.html @@ -356,7 +356,7 @@

Object Detection - + diff --git a/categories/Pedestrian-Detection/index.html b/categories/Pedestrian-Detection/index.html index 547d33d..386cf20 100644 --- a/categories/Pedestrian-Detection/index.html +++ b/categories/Pedestrian-Detection/index.html @@ -333,7 +333,7 @@

Pedestrian Detection - + diff --git a/categories/Python/index.html b/categories/Python/index.html index 6d85d2f..6204048 100644 --- a/categories/Python/index.html +++ b/categories/Python/index.html @@ -333,7 +333,7 @@

Python - + diff --git a/categories/Tools/index.html b/categories/Tools/index.html index df88a3a..4e59038 100644 --- a/categories/Tools/index.html +++ b/categories/Tools/index.html @@ -290,7 +290,7 @@

Tools - + diff --git a/categories/index.html b/categories/index.html index 814fc92..9443deb 100644 --- a/categories/index.html +++ b/categories/index.html @@ -285,7 +285,7 @@

categories - + diff --git a/css/main.css b/css/main.css index 17aba23..af262d0 100644 --- a/css/main.css +++ b/css/main.css @@ -1411,7 +1411,7 @@ pre code { vertical-align: middle; } .links-of-author a::before { - background: #ff5b1d; + background: #99582e; display: inline-block; margin-right: 3px; transform: translateY(-2px); diff --git a/index.html b/index.html index b025e44..bd64b4d 100644 --- a/index.html +++ b/index.html @@ -1153,7 +1153,7 @@

- + diff --git a/page/2/index.html b/page/2/index.html index b5a0e8b..1891037 100644 --- a/page/2/index.html +++ b/page/2/index.html @@ -1190,7 +1190,7 @@

- + diff --git a/page/3/index.html b/page/3/index.html index e73c7d9..495ab19 100644 --- a/page/3/index.html +++ b/page/3/index.html @@ -426,7 +426,7 @@

- + diff --git a/search.xml b/search.xml index cb09d5b..5866bcc 100644 --- a/search.xml +++ b/search.xml @@ -254,34 +254,6 @@ Beancount 打造个人的记账系统

Tools - - 变量——看见社会小趋势 - /2019/06/20/Book-BianLiang/ - 作者简介

-

何帆，男，现任北京大学汇丰商学院经济学教授，兼熵一资本首席经济学家。曾任中国社会科学院世界经济与政治研究所副所长，在政策研究领域研究已经超过20年 -[3] ，发表学术论文100多篇，出版专著10余部，如《变量》《何帆大局观》等 -。现被厦门大学EMBA管理学院特聘为EMBA讲师，同陆磊教授一同讲授EMBA课程-《宏观经济理论与实践》。同时，何帆是得到App《何帆大局观》《何帆的读书俱乐部》《何帆报告》课程主理人。

-

作者简介来自百度百科

-

摘抄

-

第一章　这样观察一棵树

-

2018年是一个新的开端。生活在2018年的人感受到的是中国经济遇到的各种冲击：中美贸易战、经济增长回落、股市下跌。他们会感到焦虑和担忧。旧路标已经消失，新秩序尚未出现。未来30年出现的一系列变化将挑战我们的认知，但历史从来都是一位“魔术师”，未来会出现意想不到的变化。在这一章，我会讲述如何像细致地观察一棵树一样观察历史，怎样从每年长出的“嫩芽”去判断中国文明这棵大树的生命力。我还会告诉你两个重要的概念：慢变量和小趋势。感知历史，就要会从慢变量中寻找小趋势。

-

第二章　在无人地带寻找无人机

-

2018年，关于技术发展路径的讨论引起全民关注。中国到底是应该集中全力补上“核心技术”，还是应该扬己所长发展“应用技术”呢？我将带你回顾美国在工业革命时期的经验，并试图发现中国在信息化时代的最佳战略。我找到的第二个变量是：技术赋能。在创新阶段，寻找新技术的应用场景更重要，在边缘地带更容易找到新技术的应用场景，技术必须与市场需求匹配。我们会到新疆去看无人机，而你很可能会在酒店里邂逅机器人。中国革命的成功靠的是“群众路线”，中国经济的崛起也要走“群众路线”。

-

第三章　老兵不死

-

2018年，谁是新兴产业，谁是传统产业？哪个更胜一筹？在过去几年，互联网大军就好像当年来自中亚大草原的游牧民族，兵强马壮，来去如风。在互联网大军的攻势下，传统产业的护城河形同虚设。到了2018年，这股“为跨不破”，精于“降维打击”的大军，却在一座城堡前久攻不下。这就是工业化的代表————已经有上百年历史的汽车行业。2018年，我发现的第三个变量是：老兵不死。我要带你到传统制造业的腹地，看看他们是如何抵御互联网行业的迅猛攻势。在这里，你会看到，传统行业的老兵早已经悄悄穿上了新的军装，而新兴的产业正在积极地想传统产业学习。新兴产业和传统产业的边界，也许并没有你想象的那般泾渭分明。

-

第四章　在菜市场遇见城市设计师

-

2018年，人们最关心的是房价是否会出现拐点，但从长时间来看更值得关注的是城市化的拐点。自上而下的城市化已不可持续。我观察到的第四个变量是：自下而上的力量浮出水面。城市化的进程不会停止，未来会有更多的城市圈，但这些都市圈是放大了的城市，还是一种新的城市物种呢？未来的城市不一定都能扩张，假如城市不得不“收缩”，该怎样才能像瘦身一样，瘦了更健康？未来的城市将深受互联网影响，城市空间布局会跟过去有很大的不同。“位置、位置、位置”的传统房地产“金律”很可能不再适用。我们会看到，城市会爆发一场“颜值革命”。这场“颜值革命”来自哪里呢？归根到底，它来自人民群众自己创造美好生活的能量。

-

第五章　阿那亚和范家小学

-

2018年，我们听到了很多负面的社会新闻：米脂杀人、衡阳装车、高铁霸座......这个社会变得越来越糟糕了吗？其中这是一种误解。虽然从表明上看，有些人只关心自我私利，但大家对集体生活的向往并没有泯灭。中国人已经意识到，只有重建集体生活，才能更好地发现自我。我看到的第五个变量就是：重建社群。有哪些地方的人们正在“凝结”起来，形成新的社群？这些新的社群只是孤岛，还是将成为群岛？培养孩子也需要一个社群。我会带你到一所偏僻的农村小学看看。2018年，我找到的中国教育理念最先进的小学不是北京或上海名校，而是山区里的一所农村小学。你不必吃惊，社会发展的剧情经常会有令人意想不到的转变。

-]]> - - Book - - - Book - - Deep High-Resolution Representation Learning for Human Pose Estimation /2019/05/23/CVPR19-HRNet/ @@ -358,6 +330,34 @@ unit）输出的高分辨率表示中回归热图。损失函数（定义为均 Pose Estimation + + 变量——看见社会小趋势 + /2019/06/20/Book-BianLiang/ + 作者简介 +

何帆，男，现任北京大学汇丰商学院经济学教授，兼熵一资本首席经济学家。曾任中国社会科学院世界经济与政治研究所副所长，在政策研究领域研究已经超过20年 +[3] ，发表学术论文100多篇，出版专著10余部，如《变量》《何帆大局观》等 +。现被厦门大学EMBA管理学院特聘为EMBA讲师，同陆磊教授一同讲授EMBA课程-《宏观经济理论与实践》。同时，何帆是得到App《何帆大局观》《何帆的读书俱乐部》《何帆报告》课程主理人。

+

作者简介来自百度百科

+

摘抄

+

第一章　这样观察一棵树

+

2018年是一个新的开端。生活在2018年的人感受到的是中国经济遇到的各种冲击：中美贸易战、经济增长回落、股市下跌。他们会感到焦虑和担忧。旧路标已经消失，新秩序尚未出现。未来30年出现的一系列变化将挑战我们的认知，但历史从来都是一位“魔术师”，未来会出现意想不到的变化。在这一章，我会讲述如何像细致地观察一棵树一样观察历史，怎样从每年长出的“嫩芽”去判断中国文明这棵大树的生命力。我还会告诉你两个重要的概念：慢变量和小趋势。感知历史，就要会从慢变量中寻找小趋势。

+

第二章　在无人地带寻找无人机

+

2018年，关于技术发展路径的讨论引起全民关注。中国到底是应该集中全力补上“核心技术”，还是应该扬己所长发展“应用技术”呢？我将带你回顾美国在工业革命时期的经验，并试图发现中国在信息化时代的最佳战略。我找到的第二个变量是：技术赋能。在创新阶段，寻找新技术的应用场景更重要，在边缘地带更容易找到新技术的应用场景，技术必须与市场需求匹配。我们会到新疆去看无人机，而你很可能会在酒店里邂逅机器人。中国革命的成功靠的是“群众路线”，中国经济的崛起也要走“群众路线”。

+

第三章　老兵不死

+

2018年，谁是新兴产业，谁是传统产业？哪个更胜一筹？在过去几年，互联网大军就好像当年来自中亚大草原的游牧民族，兵强马壮，来去如风。在互联网大军的攻势下，传统产业的护城河形同虚设。到了2018年，这股“为跨不破”，精于“降维打击”的大军，却在一座城堡前久攻不下。这就是工业化的代表————已经有上百年历史的汽车行业。2018年，我发现的第三个变量是：老兵不死。我要带你到传统制造业的腹地，看看他们是如何抵御互联网行业的迅猛攻势。在这里，你会看到，传统行业的老兵早已经悄悄穿上了新的军装，而新兴的产业正在积极地想传统产业学习。新兴产业和传统产业的边界，也许并没有你想象的那般泾渭分明。

+

第四章　在菜市场遇见城市设计师

+

2018年，人们最关心的是房价是否会出现拐点，但从长时间来看更值得关注的是城市化的拐点。自上而下的城市化已不可持续。我观察到的第四个变量是：自下而上的力量浮出水面。城市化的进程不会停止，未来会有更多的城市圈，但这些都市圈是放大了的城市，还是一种新的城市物种呢？未来的城市不一定都能扩张，假如城市不得不“收缩”，该怎样才能像瘦身一样，瘦了更健康？未来的城市将深受互联网影响，城市空间布局会跟过去有很大的不同。“位置、位置、位置”的传统房地产“金律”很可能不再适用。我们会看到，城市会爆发一场“颜值革命”。这场“颜值革命”来自哪里呢？归根到底，它来自人民群众自己创造美好生活的能量。

+

第五章　阿那亚和范家小学

+

2018年，我们听到了很多负面的社会新闻：米脂杀人、衡阳装车、高铁霸座......这个社会变得越来越糟糕了吗？其中这是一种误解。虽然从表明上看，有些人只关心自我私利，但大家对集体生活的向往并没有泯灭。中国人已经意识到，只有重建集体生活，才能更好地发现自我。我看到的第五个变量就是：重建社群。有哪些地方的人们正在“凝结”起来，形成新的社群？这些新的社群只是孤岛，还是将成为群岛？培养孩子也需要一个社群。我会带你到一所偏僻的农村小学看看。2018年，我找到的中国教育理念最先进的小学不是北京或上海名校，而是山区里的一所农村小学。你不必吃惊，社会发展的剧情经常会有令人意想不到的转变。

+]]> + + Book + + + Book + + High-level Semantic Feature Detection A New Perspective for Pedestrian Detection /2019/05/29/CVPR2019-CSP/ @@ -513,49 +513,6 @@ https://mirrors.tuna.tsinghua.edu.cn/anaconda/miniconda/ 下载。

Python - - CornerNet: Detection Objects as Paired Keypoints - /2018/09/02/CornerNet-Detection-Objects-as-Paired-Keypoints/ - 前言 -

CornerNet: Detection Objects as Paired -Keypoints　这篇论文发表在ECCV2018，本人感觉非常有意思，所以和大家分享一下。

-

Arxiv: https://arxiv.org/abs/1808.01244 Github: -https://github.com/umich-vl/

- --- - - - - - - - - - - - -
### 介绍
传统的目标检测都是给出紧致的候选框，本论文独具匠心，通过一对关键点（目标的左上角和右下角）来检测一个目标框。通过检测关键点的这种方式，可以消除利用先验知识设计anchor -boxes这个需求。作者提出角点池化（corner -pooling），角点池化可以帮助网络更好的定位角点。最终实验表明，CornerNet在MS -COCO数据集上实现了42.1%的AP，优于所有现存的单级(one-stage)检测器。
- - - - - - - - -]]> - - Object Detection - - - Object Detection - - Deformable DETR /2021/11/24/Deformable-DETR/ @@ -1005,66 +962,6 @@ loss主要设计用于解决one-stage检测系统中的这些问题。

Object Detection - - Generative Adversarial Nets - /2018/08/14/Generative-Adversarial-Nets/ - DCGANs in TensorFlow -

carpedm20/DCGAN-tensorflow -我们定义网络结构：

-

def generator(self, z):
    self.z_, self.h0_w, self.h0_b = linear(z, self.gf_dim*8*4*4,
                                           'g_h0_lin', with_w=True)

    self.h0 = tf.reshape(self.z_, [-1, 4, 4, self.gf_dim * 8])
    h0 = tf.nn.relu(self.g_bn0(self.h0))

    self.h1, self.h1_w, self.h1_b = conv2d_transpose(h0,
        [self.batch_size, 8, 8, self.gf_dim*4], name='g_h1', with_w=True)
    h1 = tf.nn.relu(self.g_bn1(self.h1))

    h2, self.h2_w, self.h2_b = conv2d_transpose(h1,
        [self.batch_size, 16, 16, self.gf_dim*2], name='g_h2', with_w=True)
    h2 = tf.nn.relu(self.g_bn2(h2))

    h3, self.h3_w, self.h3_b = conv2d_transpose(h2,
        [self.batch_size, 32, 32, self.gf_dim*1], name='g_h3', with_w=True)
    h3 = tf.nn.relu(self.g_bn3(h3))

    h4, self.h4_w, self.h4_b = conv2d_transpose(h3,
        [self.batch_size, 64, 64, 3], name='g_h4', with_w=True)

    return tf.nn.tanh(h4)

def discriminator(self, image, reuse=False):
    if reuse:
        tf.get_variable_scope().reuse_variables()

    h0 = lrelu(conv2d(image, self.df_dim, name='d_h0_conv'))
    h1 = lrelu(self.d_bn1(conv2d(h0, self.df_dim*2, name='d_h1_conv')))
    h2 = lrelu(self.d_bn2(conv2d(h1, self.df_dim*4, name='d_h2_conv')))
    h3 = lrelu(self.d_bn3(conv2d(h2, self.df_dim*8, name='d_h3_conv')))
    h4 = linear(tf.reshape(h3, [-1, 8192]), 1, 'd_h3_lin')

    return tf.nn.sigmoid(h4), h4

-

当我们初始化这个类时，我们将使用这些函数来创建模型。 -我们需要两个版本的鉴别器共享（或重用）参数。 -一个用于来自数据分布的图像的minibatch，另一个用于来自发生器的图像的minibatch。

-
self.G = self.generator(self.z)
self.D, self.D_logits = self.discriminator(self.images)
self.D_, self.D_logits_ = self.discriminator(self.G, reuse=True)

-

接下来我们定义损失函数。我们在D的预测值和我们理想的判别器输出值之间使用交叉熵，而没有只用求和，因为这样的效果更好。判别器希望对“真实”数据的预测全部是1，并且来自生成器的“假”数据的预测全部是零。生成器希望判别器对所有假样本的预测都是1。

-
self.d_loss_real = tf.reduce_mean(
    tf.nn.sigmoid_cross_entropy_with_logits(self.D_logits,
                                            tf.ones_like(self.D)))
self.d_loss_fake = tf.reduce_mean(
    tf.nn.sigmoid_cross_entropy_with_logits(self.D_logits_,
                                            tf.zeros_like(self.D_)))
self.d_loss = self.d_loss_real + self.d_loss_fake

self.g_loss = tf.reduce_mean(
    tf.nn.sigmoid_cross_entropy_with_logits(self.D_logits_,
                                            tf.ones_like(self.D_)))

-

收集每个模型的变量，以便可以单独进行训练。

-
t_vars = tf.trainable_variables()

self.d_vars = [var for var in t_vars if 'd_' in var.name]
self.g_vars = [var for var in t_vars if 'g_' in var.name]

-

现在我们准备好优化参数，我们将使用ADAM，这是一种在现代深度学习中常见的自适应非凸优化方法。ADAM通常与SGD竞争，并且（通常）不需要手动调节学习速率，动量和其他超参数。

-
d_optim = tf.train.AdamOptimizer(config.learning_rate, beta1=config.beta1) \
                    .minimize(self.d_loss, var_list=self.d_vars)
g_optim = tf.train.AdamOptimizer(config.learning_rate, beta1=config.beta1) \
                    .minimize(self.g_loss, var_list=self.g_vars)
-

我们已经准备好了解我们的数据。在每个epoch中，我们在每个minibatch中采样一些图像，并且运行优化器更新网络。有趣的是，如果G仅更新一次，判别器的损失则不会为零。另外，我认为d_loss_fake和d_loss_real在最后的额外的调用回到是一点点不必要的计算，并且是冗余的，因为这些值是作为d_optim和g_optim的一部分计算的。作为TensorFlow中的练习，您可以尝试优化此部分并将RP发送到原始库。

-
for epoch in xrange(config.epoch):
    ...
    for idx in xrange(0, batch_idxs):
        batch_images = ...
        batch_z = np.random.uniform(-1, 1, [config.batch_size, self.z_dim]) \
                    .astype(np.float32)

        # Update D network
        _, summary_str = self.sess.run([d_optim, self.d_sum],
            feed_dict={ self.images: batch_images, self.z: batch_z })

        # Update G network
        _, summary_str = self.sess.run([g_optim, self.g_sum],
            feed_dict={ self.z: batch_z })

        # Run g_optim twice to make sure that d_loss does not go to zero
        # (different from paper)
        _, summary_str = self.sess.run([g_optim, self.g_sum],
            feed_dict={ self.z: batch_z })

        errD_fake = self.d_loss_fake.eval({self.z: batch_z})
        errD_real = self.d_loss_real.eval({self.images: batch_images})
        errG = self.g_loss.eval({self.z: batch_z})
-

Generative Adversarial -Networks代码整理

-
    -

  • InfoGAN-TensorFlow:InfoGAN: -Interpretable Representation Learning by Information Maximizing -Generative Adversarial Nets

    • -

  • iGAN-Theano:Generative -Visual Manipulation on the Natural Image Manifold

    • -

  • SeqGAN-TensorFlow:SeqGAN: -Sequence Generative Adversarial Nets with Policy Gradient

    • -

  • DCGAN-Tensorflow:Deep -Convolutional Generative Adversarial Networks

    • -

  • dcgan_code-Theano:Unsupervised -Representation Learning with Deep Convolutional Generative Adversarial -Networks

    • -

  • improved-gan-Theano:Improved -Techniques for Training GANs

    • -

  • chainer-DCGAN:Chainer -implementation of Deep Convolutional Generative Adversarial -Network

    • -

  • keras-dcgan

    • -
-]]> - - Deep Learning - - - GAN - - How to use hexo? /2018/08/14/How-to-use-hexo/ @@ -1174,6 +1071,179 @@ href="https://www.tzingtao.com/2019/06/21/Math-Equations-in-Hexo/">Rendering Math Equations in Hexo

]]> + + Generative Adversarial Nets + /2018/08/14/Generative-Adversarial-Nets/ + DCGANs in TensorFlow +

carpedm20/DCGAN-tensorflow +我们定义网络结构：

+

def generator(self, z):
    self.z_, self.h0_w, self.h0_b = linear(z, self.gf_dim*8*4*4,
                                           'g_h0_lin', with_w=True)

    self.h0 = tf.reshape(self.z_, [-1, 4, 4, self.gf_dim * 8])
    h0 = tf.nn.relu(self.g_bn0(self.h0))

    self.h1, self.h1_w, self.h1_b = conv2d_transpose(h0,
        [self.batch_size, 8, 8, self.gf_dim*4], name='g_h1', with_w=True)
    h1 = tf.nn.relu(self.g_bn1(self.h1))

    h2, self.h2_w, self.h2_b = conv2d_transpose(h1,
        [self.batch_size, 16, 16, self.gf_dim*2], name='g_h2', with_w=True)
    h2 = tf.nn.relu(self.g_bn2(h2))

    h3, self.h3_w, self.h3_b = conv2d_transpose(h2,
        [self.batch_size, 32, 32, self.gf_dim*1], name='g_h3', with_w=True)
    h3 = tf.nn.relu(self.g_bn3(h3))

    h4, self.h4_w, self.h4_b = conv2d_transpose(h3,
        [self.batch_size, 64, 64, 3], name='g_h4', with_w=True)

    return tf.nn.tanh(h4)

def discriminator(self, image, reuse=False):
    if reuse:
        tf.get_variable_scope().reuse_variables()

    h0 = lrelu(conv2d(image, self.df_dim, name='d_h0_conv'))
    h1 = lrelu(self.d_bn1(conv2d(h0, self.df_dim*2, name='d_h1_conv')))
    h2 = lrelu(self.d_bn2(conv2d(h1, self.df_dim*4, name='d_h2_conv')))
    h3 = lrelu(self.d_bn3(conv2d(h2, self.df_dim*8, name='d_h3_conv')))
    h4 = linear(tf.reshape(h3, [-1, 8192]), 1, 'd_h3_lin')

    return tf.nn.sigmoid(h4), h4

+

当我们初始化这个类时，我们将使用这些函数来创建模型。 +我们需要两个版本的鉴别器共享（或重用）参数。 +一个用于来自数据分布的图像的minibatch，另一个用于来自发生器的图像的minibatch。

+
self.G = self.generator(self.z)
self.D, self.D_logits = self.discriminator(self.images)
self.D_, self.D_logits_ = self.discriminator(self.G, reuse=True)

+

接下来我们定义损失函数。我们在D的预测值和我们理想的判别器输出值之间使用交叉熵，而没有只用求和，因为这样的效果更好。判别器希望对“真实”数据的预测全部是1，并且来自生成器的“假”数据的预测全部是零。生成器希望判别器对所有假样本的预测都是1。

+
self.d_loss_real = tf.reduce_mean(
    tf.nn.sigmoid_cross_entropy_with_logits(self.D_logits,
                                            tf.ones_like(self.D)))
self.d_loss_fake = tf.reduce_mean(
    tf.nn.sigmoid_cross_entropy_with_logits(self.D_logits_,
                                            tf.zeros_like(self.D_)))
self.d_loss = self.d_loss_real + self.d_loss_fake

self.g_loss = tf.reduce_mean(
    tf.nn.sigmoid_cross_entropy_with_logits(self.D_logits_,
                                            tf.ones_like(self.D_)))

+

收集每个模型的变量，以便可以单独进行训练。

+
t_vars = tf.trainable_variables()

self.d_vars = [var for var in t_vars if 'd_' in var.name]
self.g_vars = [var for var in t_vars if 'g_' in var.name]

+

现在我们准备好优化参数，我们将使用ADAM，这是一种在现代深度学习中常见的自适应非凸优化方法。ADAM通常与SGD竞争，并且（通常）不需要手动调节学习速率，动量和其他超参数。

+
d_optim = tf.train.AdamOptimizer(config.learning_rate, beta1=config.beta1) \
                    .minimize(self.d_loss, var_list=self.d_vars)
g_optim = tf.train.AdamOptimizer(config.learning_rate, beta1=config.beta1) \
                    .minimize(self.g_loss, var_list=self.g_vars)
+

我们已经准备好了解我们的数据。在每个epoch中，我们在每个minibatch中采样一些图像，并且运行优化器更新网络。有趣的是，如果G仅更新一次，判别器的损失则不会为零。另外，我认为d_loss_fake和d_loss_real在最后的额外的调用回到是一点点不必要的计算，并且是冗余的，因为这些值是作为d_optim和g_optim的一部分计算的。作为TensorFlow中的练习，您可以尝试优化此部分并将RP发送到原始库。

+
for epoch in xrange(config.epoch):
    ...
    for idx in xrange(0, batch_idxs):
        batch_images = ...
        batch_z = np.random.uniform(-1, 1, [config.batch_size, self.z_dim]) \
                    .astype(np.float32)

        # Update D network
        _, summary_str = self.sess.run([d_optim, self.d_sum],
            feed_dict={ self.images: batch_images, self.z: batch_z })

        # Update G network
        _, summary_str = self.sess.run([g_optim, self.g_sum],
            feed_dict={ self.z: batch_z })

        # Run g_optim twice to make sure that d_loss does not go to zero
        # (different from paper)
        _, summary_str = self.sess.run([g_optim, self.g_sum],
            feed_dict={ self.z: batch_z })

        errD_fake = self.d_loss_fake.eval({self.z: batch_z})
        errD_real = self.d_loss_real.eval({self.images: batch_images})
        errG = self.g_loss.eval({self.z: batch_z})
+

Generative Adversarial +Networks代码整理

+
    +

  • InfoGAN-TensorFlow:InfoGAN: +Interpretable Representation Learning by Information Maximizing +Generative Adversarial Nets

    • +

  • iGAN-Theano:Generative +Visual Manipulation on the Natural Image Manifold

    • +

  • SeqGAN-TensorFlow:SeqGAN: +Sequence Generative Adversarial Nets with Policy Gradient

    • +

  • DCGAN-Tensorflow:Deep +Convolutional Generative Adversarial Networks

    • +

  • dcgan_code-Theano:Unsupervised +Representation Learning with Deep Convolutional Generative Adversarial +Networks

    • +

  • improved-gan-Theano:Improved +Techniques for Training GANs

    • +

  • chainer-DCGAN:Chainer +implementation of Deep Convolutional Generative Adversarial +Network

    • +

  • keras-dcgan

    • +
+]]> + + Deep Learning + + + GAN + + + + CornerNet: Detection Objects as Paired Keypoints + /2018/09/02/CornerNet-Detection-Objects-as-Paired-Keypoints/ + 前言 +

CornerNet: Detection Objects as Paired +Keypoints　这篇论文发表在ECCV2018，本人感觉非常有意思，所以和大家分享一下。

+

Arxiv: https://arxiv.org/abs/1808.01244 Github: +https://github.com/umich-vl/

+ +++ + + + + + + + + + + + +
### 介绍
传统的目标检测都是给出紧致的候选框，本论文独具匠心，通过一对关键点（目标的左上角和右下角）来检测一个目标框。通过检测关键点的这种方式，可以消除利用先验知识设计anchor +boxes这个需求。作者提出角点池化（corner +pooling），角点池化可以帮助网络更好的定位角点。最终实验表明，CornerNet在MS +COCO数据集上实现了42.1%的AP，优于所有现存的单级(one-stage)检测器。
+ + + + + + + + +]]> + + Object Detection + + + Object Detection + + + + Keras Tutorial + /2018/08/14/Keras-Tutorial/ + Github地址：here

+

Keras-Tutorials

+
+

版本：0.0.1

+
+
+

作者：张天亮

+
+
+

邮箱：zhangtianliang13@mails.ucas.ac.cn

+
+

Github 加载 .ipynb 的速度较慢，建议在 Nbviewer +中查看该项目。

+

简介

+

大部分内容来自keras项目中的examples

+

目录

+ +

更多Keras使用方法请查看手册 - 中文手册 - 英文手册 - github

+]]> + + Deep Learning + + + Keras + + K-Net:Towards Unified Image Segmentation /2022/05/20/K-Net/ @@ -1447,76 +1517,6 @@ after 10 iterations the average error rate is: 0.379104 Machine Learning - - Keras Tutorial - /2018/08/14/Keras-Tutorial/ - Github地址：here

-

Keras-Tutorials

-
-

版本：0.0.1

-
-
-

作者：张天亮

-
-
-

邮箱：zhangtianliang13@mails.ucas.ac.cn

-
-

Github 加载 .ipynb 的速度较慢，建议在 Nbviewer -中查看该项目。

-

简介

-

大部分内容来自keras项目中的examples

-

目录

- -

更多Keras使用方法请查看手册 - 中文手册 - 英文手册 - github

-]]> - - Deep Learning - - - Keras - - Pandas Tutorial /2019/06/23/Pandas-Tutorial/ diff --git a/tags/Beancount/index.html b/tags/Beancount/index.html index 6b68c4a..713cc75 100644 --- a/tags/Beancount/index.html +++ b/tags/Beancount/index.html @@ -290,7 +290,7 @@

Beancount - + diff --git a/tags/Book/index.html b/tags/Book/index.html index 79674d0..eb57a08 100644 --- a/tags/Book/index.html +++ b/tags/Book/index.html @@ -290,7 +290,7 @@

Book - + diff --git a/tags/Deep-Learning/index.html b/tags/Deep-Learning/index.html index 4eaed25..deacc3c 100644 --- a/tags/Deep-Learning/index.html +++ b/tags/Deep-Learning/index.html @@ -370,7 +370,7 @@

Deep Learning - + diff --git a/tags/GAN/index.html b/tags/GAN/index.html index bb75606..added2f 100644 --- a/tags/GAN/index.html +++ b/tags/GAN/index.html @@ -290,7 +290,7 @@

GAN - + diff --git a/tags/Instance-Segmentation/index.html b/tags/Instance-Segmentation/index.html index 9c65d09..1d3c985 100644 --- a/tags/Instance-Segmentation/index.html +++ b/tags/Instance-Segmentation/index.html @@ -310,7 +310,7 @@

Instance Segmentation - + diff --git a/tags/Keras/index.html b/tags/Keras/index.html index c34685b..9d503e5 100644 --- a/tags/Keras/index.html +++ b/tags/Keras/index.html @@ -290,7 +290,7 @@

Keras - + diff --git a/tags/Object-Detection/index.html b/tags/Object-Detection/index.html index 3841bff..72b7ede 100644 --- a/tags/Object-Detection/index.html +++ b/tags/Object-Detection/index.html @@ -356,7 +356,7 @@

Object Detection - + diff --git a/tags/Pandas/index.html b/tags/Pandas/index.html index 1614333..49701a8 100644 --- a/tags/Pandas/index.html +++ b/tags/Pandas/index.html @@ -290,7 +290,7 @@

Pandas - + diff --git a/tags/Pedestrian-Detection/index.html b/tags/Pedestrian-Detection/index.html index 86b031b..874a54f 100644 --- a/tags/Pedestrian-Detection/index.html +++ b/tags/Pedestrian-Detection/index.html @@ -333,7 +333,7 @@

Pedestrian Detection - + diff --git a/tags/Pose-Estimation/index.html b/tags/Pose-Estimation/index.html index e8f4ffb..1385e59 100644 --- a/tags/Pose-Estimation/index.html +++ b/tags/Pose-Estimation/index.html @@ -290,7 +290,7 @@

Pose Estimation - + diff --git a/tags/Python/index.html b/tags/Python/index.html index 2527f99..522f7db 100644 --- a/tags/Python/index.html +++ b/tags/Python/index.html @@ -310,7 +310,7 @@

Python - + diff --git a/tags/Tools/index.html b/tags/Tools/index.html index 4445e64..916135f 100644 --- a/tags/Tools/index.html +++ b/tags/Tools/index.html @@ -290,7 +290,7 @@

Tools - + diff --git a/tags/Transformer/index.html b/tags/Transformer/index.html index 76bb865..f0040cc 100644 --- a/tags/Transformer/index.html +++ b/tags/Transformer/index.html @@ -333,7 +333,7 @@

Transformer - + diff --git a/tags/docker/index.html b/tags/docker/index.html index 36fb1ce..04f01ce 100644 --- a/tags/docker/index.html +++ b/tags/docker/index.html @@ -290,7 +290,7 @@

docker - + diff --git a/tags/index.html b/tags/index.html index d8b1d0b..716d035 100644 --- a/tags/index.html +++ b/tags/index.html @@ -285,7 +285,7 @@

tags - +