Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Question about Build_conv_block #1

Open
sonnguyen129 opened this issue Jan 5, 2022 · 7 comments
Open

Question about Build_conv_block #1

sonnguyen129 opened this issue Jan 5, 2022 · 7 comments

Comments

@sonnguyen129
Copy link

When I read your code, I see this function
def build_conv_block(self, dim, padding_type, norm_layer, use_dropout, use_bias, cated_stream2=False, cal_att=False)

Can you explain two params cated_stream2 and cal_att? Thank you so much.

@sonnguyen129
Copy link
Author

Hi @Amazingren, please explain to me. Thank you so much.

@Amazingren
Copy link
Owner

Hi @sonnguyen129 ,

Thanks for the interest in our work, also sorry for the late reply since we are hurried on another project.
The function build_conv_block is just the code used in our previous projects, the two params cated_stream2 and cal_att in the CrossMLP projects are just redundant and won't affect the performance of the experiments, please ignore them. Sorry for bringing you the confusion.

Best regards
Bin Ren

@sonnguyen129
Copy link
Author

Hi @Amazingren
Thanks for your reply. Can you explain why use Bacthnorm in build_conv_block and use_bias = InstanceNorm? Hope to hear from you soon.

@sonnguyen129
Copy link
Author

Hi @Amazingren
I have another question. What if I use input F_s is a normal feature map instead of a semantic map in Cross MLP-Mixer block?

@Amazingren
Copy link
Owner

Hi @Amazingren Thanks for your reply. Can you explain why use Bacthnorm in build_conv_block and use_bias = InstanceNorm? Hope to hear from you soon.
Hi @sonnguyen129.

That's used to figure out which kind of normalization method will boot the performance of our method. And we found InstansNorm here is better than. But it depends, you can also modify them to do further exploration.

@Amazingren
Copy link
Owner

Hi @Amazingren I have another question. What if I use input F_s is a normal feature map instead of a semantic map in Cross MLP-Mixer block?

I think it is possible to take F_s as a normal feature based on your task. The direct thing you need to do maybe is just to adjust give some small modifications to the first block, for example, change the channel numbers of the first convolutional layers.

@sonnguyen129
Copy link
Author

Hi @Amazingren
Thank you for your reply. Is the first block you mentioned Conv blocks circled in red?
image

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants