"Broken" plinth, because of background?

[LiNo]

Hi, I'm starting a couple of projects but only on a hobbyist basis. A number of the project seem to have split the main object in strange ways. For example, with the attached file set, it is a model (or an attempt at a model) of a single, stone, plinth on which a number of (flat but coloured / textured) objects sit. For some reason (although I have masked the images to hide the fence which is in the background) the fence appears and the plinth seems to split into two different parts in the foreground. Any idea why this is likely to happen and how to prevent it? I would have thought (hoped) that this is a perfect trial / learning object because of the number of patterns to match with.

Thanks for your help in advance,

Noel.

[LiNo]

(Apologies, I just saw that the uploads didn't make it! I did try to upload them again, but got another error. Is this because of the size of the file(s)? Do you only need the ZEP file (that's 238Mb - and I got errors trying to upload that as well? Is there another file that I should be uploading?)

[Andrea Alessi]

Hi LiNo,

having both Zep and images/mask is usually the best idea, so we have a complete view of what's happening.

You can either upload it to a sharing service (i.e. dropbox) and post the public link, or if you wish, i can contact you in private and we can use other means to share data

[LiNo]

Thanks Andrea, I'll upload it to onedrive and share the link with you (it is just a personal project, not a secret )

[LiNo]

Hi Andrea, The following is the link to the material. By the way, the model in question is "Plinth" (there are 4 zep files in the directory - all of which are still works in progress). The "plinth.zep" is only generated as far as a dense point cloud (but it still shows the split that I am talking about). I did generate the full textured mesh (one more than one occasion) to see if this would somehow fix the problem, but the same split that is visible in the final textured mesh is (naturally) visible here in the dense point cloud.

From my perspective, there are really two questions / issues that I am trying to resolve:
1. Why does the plinth split as it does?
2. Why does the fence still appear in the model, even though I was careful to remove any trace of it with masks from the originating photos?

https://1drv.ms/f/s!AgdskftNUq2xi5EnpF7Y_IthNuQeHQ

Again, thanks again for all of you assistance.

Noel.

(P.S. This link was not recognised by the "add a link" function so I pasted it into the body of the email. If you are having difficulties with it and can provide me with an email address, I can email the link to you (from onedrive) instead. Just let me know.)

[Andrea Alessi]

Hi Noel,

1. the plinth looks split simply because zephyr wrongly computed some camera positions. This is why it's important to always check the cameras after the sparse point cloud phase. Either remove the cameras that have not been oriented correclty or re-run the SfM phase with different settings. When you see these kind of "splits" it means the SfM phase was not entirely successful and either the images are not good enough or you don't have enough images (or both, or the subject is difficult, e.g. with reflective surfaces)

I see that you used "exhaustive". Exhaustive shuld be use very, very rarely. Most likely there is something wrong with the dataset if default or deep does not orient correctly most of the photos. In your case, most of your images are very dark

In my opinion aboyt your images: iso 100 is a good choice, however, did you use a tripod? Shooting at 1/125 may still give some micromovement. Since you're shooting with f/22 your images are too dark for those settings. Since you have the .arw, try getting crisper images with your camera software (then i suggest you export in jpg, as zephyr will load faster from jpg rather than from arw). Be careful however, as post processing is often a bad idea. Since you have the raw images, you may get something better out of them, but most likely, the more you brighten them, the more noise will be evident.

2. looking at plinth.zep, you can see that you forgot to check "use masks" in the project wizard: it happens! If you double click a camera in the "Cameras" section in the "Project" tab, you should see a "show masked" option. If not, it means the mask was not used. You must check "use masks" in the project wizard or otherwise zephyr will not load mask files.

Hope this helps!

[LiNo]

Thanks again Andrea. I appreciate what you are saying about the aperture setting and when I am in the location next, I will retake the photos' with a smaller f-stop to brighten them. I hear what you are saying about converting the raw images and will look at that as well (they seem reasonably bright in Xephyr itself but I will try brightening them in Photoshop and then saving them as JPG's) and I appreciate the need for care when making changes to the raw files.

The primary reason that I was using "exhaustive" was that this seemed to get the greatest number of cameras aligned - plus I was leaving it run overnight, so the extra time requirement was not a problem. I did rerun the workflow with Deep / high detail ... and made sure that the mask box was checked , and it does look much better. Quick follow up: I would have assumed that the higher level of detail would have given ascending levels of results (i.e. good / better / best), but is this not always the case and so there is some tuning of the parameters required in order to get the optimum results. Is this correct?

Again, thanks for all of your help with this.

Regards,

Noel.

[Andrea Alessi]

Hi Noel,

you are correct. Not always "higher" settings will yield better results.

Especially, in the SfM step, if the cameras are correctly oriented even in fast, you will not get any improvements with higher settings.

Things get trickier with other phases, as higher settings usually mean higher discretization/resolution. Ultimately, lets say the for dense/mesh/texture this is true, but only if the quality of the image is very good. If there are some images that are very blurred, or that are not ideal, you may get better result with lower settings.

Happy to help, always feel free to ask

[LiNo]

Understood. Thanks again Andrea.