[question] Store (x, y, z) of every point in every input image

[Osamabu]

I want to use meshroom to create a point cloud system. I want also to store the mapping of each point in the images and where it maps in the point cloud. This is basically the (x, y, z) coordinates of the matched point. That is if an image point is (u, v), it corresponds to (x, y, z) in the point cloud. Can I do that?
Thanks

[simogasp]

The output file of the Structure From Motion contains this information stored in Observations. This is a list of 3D points (of the cloud) and, for each of them, there is a list of 2D points indexed by the viewID which corresponds to the 2D feature point on that image.
You can extract the information from the sfm.abc you find on the node or, if you need a simpler format to process you can add a ConvertSfMFormat node to convert it:

just select the format (eg json) in the combo box and run the node.

[Osamabu]

The output file of the Structure From Motion contains this information stored in Observations. This is a list of 3D points (of the cloud) and, for each of them, there is a list of 2D points indexed by the viewID which corresponds to the 2D feature point on that image.
You can extract the information from the sfm.abc you find on the node or, if you need a simpler format to process you can add a ConvertSfMFormat node to convert it:
just select the format (eg json) in the combo box and run the node.

Thank you very much. Is it possible to do this from command line too? Thanks!

[natowi]

Yes

[simogasp]

if you mean Meshroom's command line, yes, but you need first to modify the graph as in the picture and then run only the node using bin/meshroom_compute.

You can do it also from the command line using the AliceVision binary with the same name, aliceVision_convertSfMFormat (.exe if u are on windoze).

[Osamabu]

Thank you. This was very helpful. One more question. Are these absolute measurements or relative. What I mean can I do distance measurements using these files?
Thanks

[natowi]

As far as I know, you need to scale the point cloud to a known scale first.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[stale]

This issue is closed due to inactivity. Feel free to re-open if new information is available.

[lwndlnd]

Does anyone know how to do this for every 3D position instead of only the feature points? I've tried with the Dense SfM data but it does not give a meaningful value for the image point (u, v) (see #1970)

[natowi]

@lwndlnd Does the ConvertSfMFormat node help? Structure should include depth information, using Image Whitelist you can even limit the images to export:

[lwndlnd]

@natowi Thanks for the response, the white list is actually very useful! I'm already using the ConvertSfMFormat node. My question is if this is also possible with the dense SfM data

It contains both the 3D and 2D points but the 2D points don't seem to make any sense:

Left is the SfMData directly from the structure from motion and right from the Meshing node

[natowi]

@lwndlnd When I remember correctly, you need to select the describer used before to include it in the output. In your case, select dspsift

BTW: A converter to convert depth maps files (exr) to obj meshes is wip: alicevision/AliceVision#1417

[lwndlnd]

@natowi I set the describer type to dspsift in the first ConvertSfMFormat node. If I select dspsift in the second one after the meshing no observations are exported.

I'll definitely try the depth map converter when it's ready!