Shao-Feng Zeng
Shao-Feng Zeng
I have the same problem (2.1.29 works for me) test env: windows 10, anaconda5.2,
> How to train with Superpoint instead of Superpoint_BN？ Set ```using_bn: false``` in *.yaml files
> Hello, thank you for your answer, when was the last time there is a problem is that I use your training network training superpoint successful model, applied directly to...
OK, thanks
> Hi 是完整的可以用 谢谢您的回复
Here is a pytorch version that has implemented most of the related functions [https://github.com/shaofengzeng/MySuperPoint](url) You can finish your work based on this code
Hi, H is used for coordinate transformation （randomly） of feature points, and we only need to know the location of the feature points~
According to my experience, these parameters ``` solver: loss: positive_margin: 1.0 negative_margin: 0.2 lambda_d: 0.05 lambda_loss: 10000 base_lr: 0.001 #0.0001 ``` , are the key parameters that affect descriptor loss,...