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

Details on creating a custom dataset for training. #9

Closed
1995YogeshSharma opened this issue Apr 27, 2021 · 1 comment
Closed

Details on creating a custom dataset for training. #9

1995YogeshSharma opened this issue Apr 27, 2021 · 1 comment

Comments

@1995YogeshSharma
Copy link

Hi @UditSinghParihar,

I have gone through the Readme and instructions work well for training on given dataset.

I am looking to generate a custom dataset and train on that. Can you please add more detail on the same (like how to annotate, format of data, do we need to modify some code etc)?

Thanks!

@UditSinghParihar
Copy link
Owner

Hi,

  1. If you want to provide custom correspondences, then you should change self.getGrid function in our data loader. This function basically returns ground-truth correspondences for the image pair. Here our correspondences are related by a homography, but it can also be related via some other way like warping using depth and poses, as done originally in D2Net.
  2. If you want to change our data loader, then you should obtain pos1 key points using the same upscale_positions function as currently there inside self.getGrid function, while pos2 would be generated using your way of obtaining correspondences for pos1, in our case we are warping pos1 using ground-truth homography to generate pos2.

Regards,
Udit

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