got curious about how the 3d reconstruction works when more images get added - maybe some mapillary expert can clarify
the 3d reconstruction can connect images from different sequences. but can image from one sequence help to connect two other images from another sequence ?
for example, let’s say there’s sequence A with images taken 20m apart and these images are not detected to be connected. sequence B is added with an image inbetween, and this image is detected to be connected with both images in A.
will both images in A now be connected in the 3d reconstruction ?
We try to connect every image to every image in the surrounding independently of which sequence they belong to. So if two images from a sequence are not connected to each other, they can still be connected by new images being connected to both of them. The idea is that 3d should get better and better as more images are added
thank you for the answer. just to make sure i got this right… let’s say after having sequences A and B, i view the first image in sequence A. if i just move forward, it would move to image from sequence B, then to the second image from sequence A.
if i lock sequence, will it know the connection (do the 3d transition ) from the first A image to the second A image ?
Mapillary uses all it’s photos to build up a 3d model. The 3d model is used to make transitions between photos, to improve position and photo direction. So if there are mmany photos in a street you will get a detailed 3D model, which makes it easier to correct the gps coordinates in your photos and the 3d model also makes it possible to make smooth 3d transitions between your photos.
So yes, photos from other sequences also improve your sequence, even if they are not shown in your sequence.
Did I explain this correct @Peter ?
I make mostly connected pictures. It would be interesting to examine if they connect well to the unconnected pictures made by the other mly photographers.
Yes,
this is very well explained @Harry. So, right now the pointcloud is mostly used to improve data and navigation, not to build up centimeter-accurate models. However, given good GPS and Direction, you can (we are working on that) even measure in the pointcloud.
Yes, new images that share points with others will help improve their accuracy.
) from the first A image to the second A image ?