【Feature】feature depth constrain sfm to reduce drift and recover scale

[yuancaimaiyi]

I implemented the feature depth to constrain sfm(not only the drift has been reduced, but the scale has also been recovered), and the result is as follows. The depth can be obtained through binocular patchmatch or TOF sensors!
total images: 4000 images
(1) pure colmap deal with indoor data, you can see that the result is very poor, only more than 200 registered

(2) feature depth constrain sfm, more than 800 registered . Simultaneously,reducing dirft and recovering scale !!!!!!!!

[ahojnnes]

Thanks for sharing these results. Are you able to share the code to produce these results in a fork of COLMAP or would you be willing to work on integrating them into the main COLMAP codebase?