search

naver / r2d2

Other
461 stars 86 forks source link

Generate image pairs using random homography #17

Closed hanyunlong233 closed 4 years ago

hanyunlong233 commented 4 years ago

hello! I want to use my own data to fine-tune your model to better fit our scene. I need to build my own training data set. According to the random homography mentioned in your article, could you please give me some advice?

humenbergerm commented 4 years ago

Hi!

Have you already tried the ones we suggested in the training code? We tried to find good settings for the task of visual localization where the features should be robust (not invariant) against viewpoint changes. I would start with this.

Best, Martin

hanyunlong233 commented 4 years ago

Thank you for your reply.I've already started my fine tuning.I think your model is very good, but my data is very special and needs to be fine-tuned.Thank you very much for your advice.