Thank you for the excellent work. I found the initialization strategy very interesting, especially the use of geometric consistency for filtering matched points.
Did you also try using photometric consistency between the two views for match filtering, such as comparing colors or local patches after reprojection? If so, how did it affect the initialization and final reconstruction quality? If not, was it mainly due to robustness concerns under exposure changes, occlusions, or non-Lambertian effects?
Thank you for the excellent work. I found the initialization strategy very interesting, especially the use of geometric consistency for filtering matched points.
Did you also try using photometric consistency between the two views for match filtering, such as comparing colors or local patches after reprojection? If so, how did it affect the initialization and final reconstruction quality? If not, was it mainly due to robustness concerns under exposure changes, occlusions, or non-Lambertian effects?