Attempting to reconstruct a 2 meter high, 25.4 cm diameter cylinder

[Scott Nebeker]

THIS, right next to my lens wherever I needed it. Those lights are a [em]very[/em] useful piece of kit but also [em]very[/em] pricey. Being able to bend the light source around a rounded subject is an incredible thing.

Something that might bring you to a screaming hault on this multi-camera setup is the number of cameras the plugin supports. Version 2.7 of Zephyr only allowed 2 cameras for the Lite version. I'm unsure if that restriction has been done away with or not. Given the restriction on GPU cores in the Lite version, 2 cameras might be the nice and stable amount. You'll find more information here: https://www.3dflow.net/remote-dslr-sync-shooting-plugin-download-3df-zephyr/

[Roberto]

I've downloaded the dataset and I'll run some tests tomorrow. From a quick look at the images, the problem might be related to the repeated pattern. I'll run some tests tomorrow and let you know.

[Modeler17]

Ok, Thank you so much, Roberto.

[Modeler17]

I have long used pocket wizards to synchronize shutter release with off camera lighting. They might possibly help me synchronize shutter releases across cameras.
Lots to think about and learn. Thank you, Scott.

[Roberto]

I've made some tests and the repeating pattern is definitively causing problems in this case.

However, you can still obtain a good results by tuning the parameters properly.
Starting from the preset 'close range - deep', I switched to advanced properties and I've set the parameters as follows:

[img src="http://i.imgur.com/yYaZ1qF.jpg" alt="Settings" /]

Specifically, I changed two parameters. The photo ordering has been set to 'circular' for the reasons you already mentioned.
Additionally, I've also decreased the 'matching stage depth' to 'Low'.
Usually you don't want to set this parameter to 'low' but in this particular case it helps to avoid ambiguities.
Suppose Zephyr is try to match features for the image N. Since you set the photo ordering to circular, Zephyr will try to look for neighbors of the images N in the interval [N-k ... N+k].
Setting the matching stage depth to 'low', will actually keep the 'k' parameter (the search range) to a value of 2. This means that, for example, potential matches for the image 3 are looked in the image 1, the image 2, the image 4 and the image 5. Zephyr does not try to match the image 3 with the image 13 or the image 14, that might be potentially very ambiguous in this specific case. I hope my explanation is quite clear.

This is the result I obtained:

[img src="http://i.imgur.com/CTPhvaq.jpg" alt="Results" /]

[Modeler17]

Roberto, These results look much better! I am going to verify that I can reproduce them on my system.

[Modeler17]

Yes, The results were reproducible on my system. I notice the image quality index indicator put slight red tinges on the balls? Is it correct to conclude that the distorted reconstruction of the balls is due to an issue with the original issue? If so, any suggestions as to what that image issue might be?

Thank you again.

[Modeler17]

Sorry I meant to say "issue with the original image", not "original issue"

[Roberto]

The balls are badly reconstructed because they are completely textureless. This is also the reason why the image quality index gives low score on the balls.

However please pay attention also to the reconstruction bounding box. I think you should manually expand the bounding box a little bit before running the densification process:
https://www.3dflow.net/technology/documents/3df-zephyr-tutorials/tutorial-managing-bounding-box/

From a the quick reconstruction test I've noticed the bounding box partially exclude the balls area.

[Modeler17]

Thank you, I tried expanding the bounding box and perhaps got a little more of the balls.