the better the sparse cloud, the better the dense cloud ?

[jupp2000]

[Andrea Alessi]

hi Jupp,

what zephyr does in the first phase is autocalibrate the cameras and place them in the scene, so to speak.

The number of points in the sparse point cloud is not directly tied to the quality of the subsequent step. What is important is that the cameras are placed correctly.

If the solution converges correctly, getting it with a fast preset or deep preset doesn't matter. In fact in some cases where your photos are bad, you may want to use less keypoints (but stronger) rather than a lot of keypoints that can increase the odds of a false match.

So if you are confident your cameras are placed correctly (if you have the full version you can easily verify that by placing control points and verifying the reprojection error on the epipolar line) you can proceed down the pipeline without worry.

If you proceed down the pipeline and see very bad results then it's likely your orientation was wrong, and you may need to redo the SfM step.

Different is for the settings of dense/mesh. In that step, in general, higher settings will yield more detailed results, although the tradeoff is computation time and that is heavily dependent on the quality of the dataset.

[jupp2000]

Thank you Andrea!

So, if i understand that, no matter if the sparse cloud consists of just 20.000 Points or 200.000 points (correct camera recognition assumend), the important steps to produce dense quality are relatively independent from the sparse cloud?

Greetings!

[cam3d]

Thank you Andrea!

So, if i understand that, no matter if the sparse cloud consists of just 20.000 Points or 200.000 points (correct camera recognition assumend), the important steps to produce dense quality are relatively independent from the sparse cloud?

Greetings!

Correct The quality of the sparse cloud (accuracy) is far more important than the density.