add data in matches for doppelganger

[awarebayes]

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

[sarlinpe]

• 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]

See https://github.com/cvg/Hierarchical-Localization/issues/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