Dely-Yu

resnet.py 的 第42行， conv1 = conv2d(inputs, num_outputs, kernel_size=**_3_**, stride=subsample, scope='Conv1') 按照论文的网络结构，conv1 = conv2d(inputs, num_outputs, kernel_size=**_1,_** stride=subsample, scope='Conv1')。 1*1卷积，之后才是3*3的卷积。

同学你好，你能提供下对聚类后的feature map做可视化的代码吗？

1

可视化是一个理解cnn行为的很好的途径，你能上传一下代码，以及你在论文中的展示例子吗？

fix some version bugs and add some test_data(picture)

what is the difference among resnet_basis, resnet_factor, and resnet_group?

2

Thanks for your good work. Will you give more description about the meaning of these three scripts? And what is the working pipeline to reproduce the results of your paper?