Some troubles about using IIC

[WeiWenQiang123456]

First of all, thank you for your contribution, sorry to bother you with the direct email, but I do have some urgent questions about IIC, I hope you can answer them for me.Questions are as follows：

1. Obtain g'(x) data in IIC through some methods such as rotation, cropping, and color change. This is to process pictures containing semantic information. It can be seen from the eye that after transformation, the image
   Objects in are essentially unchanged. While clustering non-image data using IIC. Sometimes, such as one-dimensional data in the database, it can be transformed into nxn two-dimensional matrix data, which reflects the distribution characteristics or frequency characteristics of the data, so what method should be used at this time? What about transforming the data to obtain g'(x) data? Is the data transformation method in IIC only applicable to data containing semantic information?

2. When I use MNIST in IIC for clustering, I can clearly see the loss from 0 down to -2, almost all the way down, while using own data for clustering, the loss is always It quickly reaches a value and keeps changing up and down around it, which makes me always suspect

Is there a problem with your own model? Or the loss is always 0 or a certain value. What is the problem with this situation?

3. I would like to ask what is the difference between using VGG and ResNet for unsupervised clustering. Which model is better for clustering?