Sharpness

[Nedo]

Good Morning everyone!
I have a question regarding the sharpness of a reconstructed model,
where does it starts in zephyr if you need a model as sharp as possible?
Does it matter on the first step when adding pictures, you can start default, deep or exhausting,
is there a difference in the sharpness in this 3 modes?
or does it start in the second step: Dense cloud creation,
or does it starts in the third step: Mesh creation ?

Nedo

[Andrea Alessi]

Hi Nedo!

a bit in every step. Let me elaborate:

Structure from motion

You should disregard the sparse point cloud as the output of this phase, but more on the camera pose. The purpose of the SfM is to autocalibrate the camera (internals: focal length, distortion, etc) and the to position the camera in the scene with the right externals (position, rotation). If the cameras are not oriented correctly, all the next steps down the pipeline will suffer from this

Multi view stereo

Here is a bit tricky: the dense cloud is very important, as it is the input for the mesh generation. However, not always higher settings will yield better results. For example, if your photos are not very good, you'll get much better results with 25% resolution rather than, let say, 50%. However, if the photos are well taken, usually higher settings yield a better point cloud.

Mesh generation

Depending on the photos again, here the most important aspect is the photoconsistency: this step (which is, also, tricky, depending on the photos: here are some examples https://www.3dflow.net/tutorial-poin...-mesh-filters/ ) actually moves the vertices, so it can further improve the quality. I personally like to generate a mesh with photoconsistency on, then one with photoconsistency off and try different parameters for the photoconsistency filter in order to see if i can get better results, but it tends to be somewhat time consuming



So as you can see, even if you are interested in the mesh quality, there are other factors that directly affects everything down the reconstruction pipeline.

[Nedo]

Yesterday i tested a new project, good photos, they all got found on default, just one edge looked a little bit misaligned, so i restarted in deep,
now the edge was ok, but i was thinking what will happen if i restart in exhaust mode. Thank you for all the Informations Andrea!

[Andrea Alessi]

Always happy to help

exhaustive should be used in very rare cases.

If the orientation is not correct at default but is correct at deep, then there is no reason to go further with exhaustive. Personally i find it very useful with very low quality images and with very few features!

Protip: sometimes, try switching from the global to the incremental pipeline (and viceversa) by selecting a preset and changing it to advanced, is also a good idea before going with exhaustive, which should be the last resort (also because it's very time consuming). Depending on the dataset type, one or the other can make the difference. I personally prefer the incremental one with difficult datasets, but your mileage may vary =)