Skip to content

Conversation

awarebayes
Copy link

This proposal will allow for seamless integration with the Doppelgangers repository.

@sarlinpe
Copy link
Member

  • I need more context - why are these new fields created? There is no such fields in the output of any our models.
  • I would only accept to depend on the official Doppelganger repository, not on a fork of it.
  • Since the inference of LoFTR is slow (much more than LightGlue), I don't think that it is acceptable to require running LoFTR for each image pair. Doppelganger should work on top of sparse matches, which requires retraining the model.

@awarebayes
Copy link
Author

See #308

I need more context - why are these new fields created? There is no such fields in the output of any our models.

I use them here

I would only accept to depend on the official Doppelganger repository, not on a fork of it.

Already made a PR: link

Since the inference of LoFTR is slow (much more than LightGlue), I don't think that it is acceptable to require running LoFTR for each image pair. Doppelganger should work on top of sparse matches, which requires retraining the model.

Working on it, chef

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

Successfully merging this pull request may close these issues.

2 participants