<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[tobale的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/tobale</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; tobale]]></copyright><item><title><![CDATA[论文阅读：Talking Face Generation by Adversarially Disentangled Audio-Visual Representation]]></title><link>https://blog.youkuaiyun.com/tobale/article/details/90345303</link><guid>https://blog.youkuaiyun.com/tobale/article/details/90345303</guid><author>tobale</author><pubDate>Sun, 19 May 2019 18:35:26 +0800</pubDate><description><![CDATA[转载请注明本文出处，谢谢。


论文链接：Talking Face Generation by Adversarially Disentangled Audio-Visual Representation

代码：GitHub

视频：Video



这是一篇AAAI 2019的文章。文章的题目是通过对抗分解语音和视频的表征来生成说话的人脸。

1. Introduction

一个人脸正在...]]></description><category></category></item><item><title><![CDATA[【TensorFlow】如何判断当前进行到第几个epoch了？]]></title><link>https://blog.youkuaiyun.com/tobale/article/details/86593499</link><guid>https://blog.youkuaiyun.com/tobale/article/details/86593499</guid><author>tobale</author><pubDate>Tue, 22 Jan 2019 14:00:30 +0800</pubDate><description><![CDATA[
def read_and_decode_TFRecordDataset_deblur(tfrecords_path, batch_size, epoch_num):
    dataset = tf.data.TFRecordDataset(tfrecords_path)
    dataset = dataset.map(parser_deblur).shuffle(buffer_size=1...]]></description><category></category></item><item><title><![CDATA[teamviewer13 linux 被限制解决办法]]></title><link>https://blog.youkuaiyun.com/tobale/article/details/86491398</link><guid>https://blog.youkuaiyun.com/tobale/article/details/86491398</guid><author>tobale</author><pubDate>Tue, 15 Jan 2019 13:51:44 +0800</pubDate><description><![CDATA[
由于ID依据电脑的mac地址生成, 所以需要修改电脑的mac地址来修改ID


1. 更换linux的mac地址 (http://www.linuxfly.org/post/89/)


ifconfig eth0 down
ifconfig eth0 hw ether xx:xx:xx:xx:xx:xx
ifconfig eth0 up

2. 卸载之前的teamviewer


sudo d...]]></description><category></category></item><item><title><![CDATA[【TensorFlow】测试时BN层每次输出都一样；BN均值方差训练时必须更新！]]></title><link>https://blog.youkuaiyun.com/tobale/article/details/85102773</link><guid>https://blog.youkuaiyun.com/tobale/article/details/85102773</guid><author>tobale</author><pubDate>Wed, 19 Dec 2018 17:38:47 +0800</pubDate><description><![CDATA[tf.layers.BatchNormalization有两个参数training和trainable。
training controls whether to use the training-mode batchnorm (which uses statistics from this minibatch) or inference-mode batchnorm (which uses av...]]></description><category></category></item><item><title><![CDATA[反卷积记录]]></title><link>https://blog.youkuaiyun.com/tobale/article/details/85049902</link><guid>https://blog.youkuaiyun.com/tobale/article/details/85049902</guid><author>tobale</author><pubDate>Mon, 17 Dec 2018 15:17:15 +0800</pubDate><description><![CDATA[

一共有两种反卷积：1. Full卷积， 完整的卷积可以使得原来的定义域变大（上图所示）  2. 记录pooling index，然后扩大空间，再用卷积填充。详见参考链接。

参考：https://blog.youkuaiyun.com/qq_38906523/article/details/80520950 ...]]></description><category></category></item><item><title><![CDATA[更换硬盘导致 ubuntu 16.04 紫屏]]></title><link>https://blog.youkuaiyun.com/tobale/article/details/83757081</link><guid>https://blog.youkuaiyun.com/tobale/article/details/83757081</guid><author>tobale</author><pubDate>Mon, 05 Nov 2018 21:24:37 +0800</pubDate><description><![CDATA[由于之前在/etc/fstab中设置了自动挂载硬盘, 更换硬盘后UUID不一致, 导致紫屏.

解决方法: 
Adanvced mode
	Recovery mode
	选择 root ...... (倒数第二个?)
	输入密码
	输入: mount -w -o remount /
	vi /etc/fstab
	注释掉更换的硬盘, 或者替换为新的UUID
...]]></description><category></category></item><item><title><![CDATA[【TensorFlow】训练过程中获取某个Tensor值；只有conv1和bn1存在NAN]]></title><link>https://blog.youkuaiyun.com/tobale/article/details/83620432</link><guid>https://blog.youkuaiyun.com/tobale/article/details/83620432</guid><author>tobale</author><pubDate>Thu, 01 Nov 2018 14:58:21 +0800</pubDate><description><![CDATA[1. 在训练过程中,获取某个参数Tensor的值:

获取所有Tensor的name: [tensor.name for tensor in tf.get_default_graph().as_graph_def().node]
	根据name获得Tensor: bn_gamma = sess.graph.get_tensor_by_name('bn1_audio/batch_normalizat...]]></description><category></category></item><item><title><![CDATA[Caffe 编译报错记录]]></title><link>https://blog.youkuaiyun.com/tobale/article/details/83616507</link><guid>https://blog.youkuaiyun.com/tobale/article/details/83616507</guid><author>tobale</author><pubDate>Thu, 01 Nov 2018 10:43:02 +0800</pubDate><description><![CDATA[1. 



在Makefile.config第一行加入:LINKFLAGS := -Wl,-rpath,/media/server009/seagate/liuhan/anaconda2/lib

2. /usr/local/lib/libcaffe.so



删除/usr/local/lib/libcaffe.so]]></description><category></category></item><item><title><![CDATA[论文阅读：GANimation: Anatomically-aware Facial Animation from a Single Image]]></title><link>https://blog.youkuaiyun.com/tobale/article/details/83587140</link><guid>https://blog.youkuaiyun.com/tobale/article/details/83587140</guid><author>tobale</author><pubDate>Wed, 31 Oct 2018 17:06:11 +0800</pubDate><description><![CDATA[
转载请注明本文出处，谢谢。


论文链接：GANimation: Anatomically-aware Facial Animation from a Single Image



        我们首先来看一下它的效果图。这张图是从左边的输入图像生成笑的表情。这里有一个可以调节的参数α，用于调节表情的变化程度。所以说他生成的表情不是离散的。像之前的一些方法只能生成离散数量的表情种类，这些表...]]></description><category></category></item><item><title><![CDATA[论文阅读：Deep Matching Prior Network: Toward Tighter Multi-oriented Text Detection]]></title><link>https://blog.youkuaiyun.com/tobale/article/details/80841842</link><guid>https://blog.youkuaiyun.com/tobale/article/details/80841842</guid><author>tobale</author><pubDate>Thu, 28 Jun 2018 13:35:01 +0800</pubDate><description><![CDATA[
转载请注明本文出处，谢谢。


论文链接：Deep Matching Prior Network: Toward Tighter Multi-oriented Text Detection



1. Introduction

 



    要说文字检测，那一定要先看一下目标检测。目标检测是近几年深度学习在计算机视觉领域发展最好的应用之一。
    前一段时间华盛顿大学发布YOLO V3。...]]></description><category></category></item><item><title><![CDATA[论文阅读：Knowledge Guided Disambiguation for Large-Scale Scene Classification with Multi-Resolution CNNs]]></title><link>https://blog.youkuaiyun.com/tobale/article/details/80841622</link><guid>https://blog.youkuaiyun.com/tobale/article/details/80841622</guid><author>tobale</author><pubDate>Thu, 28 Jun 2018 13:06:38 +0800</pubDate><description><![CDATA[
转载请注明本文出处，谢谢。


场景分类文献阅读目前共有两篇：

（一）论文阅读： Scene recognition with CNNs: objects, scales and dataset bias

（二）论文阅读：Knowledge Guided Disambiguation for Large-Scale Scene Classification with Multi-Resolu...]]></description><category></category></item><item><title><![CDATA[论文阅读： Scene recognition with CNNs: objects, scales and dataset bias]]></title><link>https://blog.youkuaiyun.com/tobale/article/details/80841522</link><guid>https://blog.youkuaiyun.com/tobale/article/details/80841522</guid><author>tobale</author><pubDate>Thu, 28 Jun 2018 12:49:22 +0800</pubDate><description><![CDATA[
转载请注明本文出处，谢谢。


场景分类文献阅读目前共有两篇：

（一）论文阅读： Scene recognition with CNNs: objects, scales and dataset bias

（二）论文阅读：Knowledge Guided Disambiguation for Large-Scale Scene Classification with Multi-Resolu...]]></description><category></category></item><item><title><![CDATA[论文阅读： FaceID-GAN: Learning a Symmetry Three-Player GAN for Identity-Preserving Face Synthesis]]></title><link>https://blog.youkuaiyun.com/tobale/article/details/80841361</link><guid>https://blog.youkuaiyun.com/tobale/article/details/80841361</guid><author>tobale</author><pubDate>Thu, 28 Jun 2018 12:17:31 +0800</pubDate><description><![CDATA[
转载请注明本文出处，谢谢。


论文链接：FaceID-GAN



    文章题目叫做FACEID-GAN，学习一个对称的三者对抗的GAN，来做保留身份的人脸生成。



    首先用这张图解释一下本文的目的。对于一幅输入的人脸图像，这个GAN网络可以生成多种视角，多种表情的人脸，并且可以保留这个人的身份，也就是说生成出来的人脸仍然都还是这个人。

    油管上还有一个生动的小视频，动态...]]></description><category></category></item><item><title><![CDATA[Caffe conv_layer  和  卷积操作]]></title><link>https://blog.youkuaiyun.com/tobale/article/details/78984856</link><guid>https://blog.youkuaiyun.com/tobale/article/details/78984856</guid><author>tobale</author><pubDate>Fri, 05 Jan 2018 20:16:31 +0800</pubDate><description><![CDATA[]]></description><category></category></item><item><title><![CDATA[Caffe math_functions 矩阵函数]]></title><link>https://blog.youkuaiyun.com/tobale/article/details/78984591</link><guid>https://blog.youkuaiyun.com/tobale/article/details/78984591</guid><author>tobale</author><pubDate>Fri, 05 Jan 2018 19:39:11 +0800</pubDate><description><![CDATA[详细：http://blog.youkuaiyun.com/xg123321123/article/details/53325765
摘要：http://f.dataguru.cn/thread-739475-1-1.html
1 caffe_cpu_gemm       

功能： C=alpha*A*B+beta*C        A,B,C 是输入矩阵（一维数组格式） 
2 caffe_cpu]]></description><category></category></item><item><title><![CDATA[Caffe softmax_cross_entropy_loss_layer.cpp]]></title><link>https://blog.youkuaiyun.com/tobale/article/details/78984520</link><guid>https://blog.youkuaiyun.com/tobale/article/details/78984520</guid><author>tobale</author><pubDate>Fri, 05 Jan 2018 19:28:28 +0800</pubDate><description><![CDATA[http://blog.youkuaiyun.com/u012235274/article/details/51361290]]></description><category></category></item><item><title><![CDATA[ubuntu caffe opencv pip]]></title><link>https://blog.youkuaiyun.com/tobale/article/details/70028496</link><guid>https://blog.youkuaiyun.com/tobale/article/details/70028496</guid><author>tobale</author><pubDate>Mon, 10 Apr 2017 20:20:21 +0800</pubDate><description><![CDATA[pip opencv]]></description><category></category></item><item><title><![CDATA[Ubuntu 14.04 网易源]]></title><link>https://blog.youkuaiyun.com/tobale/article/details/59064553</link><guid>https://blog.youkuaiyun.com/tobale/article/details/59064553</guid><author>tobale</author><pubDate>Wed, 01 Mar 2017 11:21:57 +0800</pubDate><description><![CDATA[Ubuntu 14.04 网易源]]></description><category></category></item><item><title><![CDATA[Ubuntu 16.04 阿里云源]]></title><link>https://blog.youkuaiyun.com/tobale/article/details/59063068</link><guid>https://blog.youkuaiyun.com/tobale/article/details/59063068</guid><author>tobale</author><pubDate>Wed, 01 Mar 2017 11:17:44 +0800</pubDate><description><![CDATA[Ubuntu 16.04 阿里云源]]></description><category></category></item></channel></rss>