Hello, this is a question about the number of channels in the convolutional layer

The paper described:
We use three-layer encoders with [64, 32, 16] channels, and three�layer decoders with [16, 32, 64] channels correspondingly.

But the code is:
def encoder1(self, x):
net = self.conv_block(x, 64)
net = self.conv_block(net, 64)
net = self.conv_block(net, 64)
return net

Did I misunderstand? Thank you

Hi, thank you very much for your great work, could you share the code on other multi-feature datasets, thanks!!

Hello, I'm very interested in your paper. Your code is only about rgb-d data set. Can you provide the code about Yale data set processing? Thank you
My Email is [email protected]

Hi, thank you very much for your great work. This work inspires me a lot. But when trying to use other datasets in this work, I encountered difficulties. Could you share the code on other multi-feature datasets, thanks a lot!!

each sample for BBCSport dataset is one-dimensional. How to train bbcsport dataset? Do you use one-dimensional convolution for training?

mtv.reconstruct function defined in model/rgbd,py takes only 3 arguments i.e. def reconstruct(self, view1, view2, lr) but in main.py when loss is being calculated in pre-train phase, there 3 views and lr given i.e loss = mtv.reconstruct(views[0], views[1], views[2], lr).

Hello,

I am very interested to your idea that employing both diversity and universality nets to find the robust latent expressions.

In the paper, some datasets with extracted features, such as Yale, are adopted. I am wondering their network structures but do not find in the paper and code. Could you please tell me more about them ?
Thank you so much !

Jiyuan Liu
[email protected]