ht014 / bgan Goto Github PK

请问cifar_KNN.npz的特征是怎么得到的呢，由VGG-19提取的吗？
另外，代码和论文似乎也不对应，噪声并不是添加的二进制，训练的一直是连续的哈希，到最后用sign函数保存哈希码。

thanks for sharing your code.I wander how long time will take to train 'BGAN.py'program with CPU？thanks

Which version of tensorflow is used in this code? I will make mistakes with version 1.13.0.

你好，我想问下您的cifar-knn文件的K近邻是使用欧氏距离来计算的吗，为什么我重新用VGG19的relu7来重新计算KNN的时候，跟您原来的KNN差异很大，而且您论文描述的是使用resnet提取的2048维特征

I run this code in 12,24,32,48 bits(epochs51)
each of them in six times
but only 12 bits can get mAP higher than the mAP shown in the paper(0.401)
In 32 bits get 0.4599 but paper say the method can get 0.531
and 48 bit get 0.4487 but paper's mAP is 0.558

Is there some setting need to change or how many epochs is prefect ?

Thanks for sharing the code.
When I run the BGAN code, I find that the pretrained vgg19-like model is crucial. So how to train the model on ImageNet since it has two branch layers fc8 and fc10?

代码中的tanh_beta能够实现论文中分段线性函数app的效果吗？或者说fc9与fc10层输出的组合，我不太明白fc10层输出的用途。

excuse me，there is an error : AttributeError: 'VariableScope' object has no attribute 'current_scope'
is that because cifar-10.mat is necessary？

Thanks very much for sharing the code and I think this is very great job. But I have some questions in understanding the code.
Firstly, what does the KL_loss mean? According to my understand, the SSE_loss in the code corresponds to the MSE_loss in the paper, but you didn't mentioned the KL_loss in the paper?
Secondly, you did't use the KL_loss and SSE_loss when updating the params, why?
Thirdly, z_x = tf.add(z_x_mean, tf.multiply(tf.sqrt(tf.exp(z_x_log_sigma_sq)), eps)) # grab our actual z, what does "grab actual z "mean?
Finally, What's the role of the function "average_gradients" @htconquer

链接: https://pan.baidu.com/s/1at2jvObSdNJ9HAzwA57-Bg 密码: 04r2 BlurryFacess [email protected] 于2020年9月29日周二 上午11:29写道：
…
Hi, 试试这个能不能下载，链接: https://pan.baidu.com/s/13ts5KcXcSETPggz26vOQmg 密码: 61cu FullMoonZhang @.*** 于2020年9月16日周三 上午8:59写道： … <#m_3539921555070459916_> Yes, we use resnet feature to construct KNN graph. You can download KNN graph from https://drive.google.com/file/d/0Bzg9TvY-s7y2WFFlc3F0T2RkalE/view?usp=sharing 嗯嗯，请问您有nus-wide划分好的数据集的地址提供下载吗？，比如网盘这些，因为我在跑您的代码的时候，也是使用的nus-wide @@.*** @.*** <https://github.com/5000>有0.6多， @ .***是否是正常的，您论文里面nus-wide给的map@N的N是整个检索集数量吗 您好，请问您的nus-wide 21个类是从哪里下载的呢？我找了好久，但只能找到官网的版本。作者的链接也过期了，所以很抱歉麻烦您 — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#9 (comment) <#9 (comment)>>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABGWUPLJZRIIRLBJQYNFDB3SGAEVJANCNFSM4JSQKLLA . Hey，the link of baidupan you shared is no longer useful, Can i get a new share? Thx a lot — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#9 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABGWUPPMPSQKUHKLOMXOS3LSIFIATANCNFSM4JSQKLLA .

谢谢您~

Originally posted by @BlurryFaced in #9 (comment)

In generator.py line 75,
self.fc6 = self.fc_layer(self.pool5, 25088, 4096, "fc6") # 25088 = ((224 / (2 ** 5)) ** 2) * 512

So the input image size is 224*224?

thanks for sharing your code,
The results of my own training map on cifar10 is 4% less than yours. May I get your trained model please?