-
Notifications
You must be signed in to change notification settings - Fork 1
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
Hello, could you provide the training code for UFEN's binary descriptors? I've tried to replicate the training code for binary descriptors but the results have been poor. Therefore, I am seeking your help. #4
Comments
Hi, Adding a stronger noise term (N) in image synthesis (separately for the paired images) enhances the performance of the descriptor. You should consider trying it. |
Thank you for your response. I will try again following your suggestions. It would be great if you could provide the "Matching loss" as well. |
I apologize for bothering you again. I've been having trouble replicating good results in the training part of binary descriptor. Could you please provide the code for the 'Matching loss' section? I appreciate it greatly. |
Hi, |
Thank you very much for your kind help. I sincerely hope that you can achieve even more brilliant academic achievements. |
No worries, you are very welcome.
|
Hello, I would like to ask how many epochs you trained the network to achieve the expected effect? Thank you. |
The proposed weights are trained over 20 epochs. Typically, performance nearly converges after 10 epochs. If additional training or more epochs are necessary, you can incorporate a small similarity loss (e.g., L2 loss) between the descriptor outputs. This helps ensure that the new model remains close to the original SuperPoint. |
No description provided.
The text was updated successfully, but these errors were encountered: