nie-lang
commented
2 years ago Increasing the search range may introduce extensive extra parameters. It's not recommended to do it.
Most deep homography methods can not work well in scenes of low overlap. If you solve this problem well, it would be appealing to see it.
From my experience, a lightweight but effective feature matching module may be helpful.
If I want to implement this method into scenes of low overlap, what's the necessary change for the code? Do I need to increase the search range or input resolution? Thanks!