Hello there
Trying out the demo from Steam with some really exciting results.
I have noticed that sometimes when I align images it doesnt recognize more than half of my imported images. And then I would re-align just in case and it will include the rest.
Is this normal? I thought aligning images is a deterministic process. At the moment it looks like if I open the folder with same images twice I almost always get differently aligned results.
Kind regards
Hi,
the alignment process is iterative. So if you don’t clear the RC cache, the next alignment will take previously aligned components into consideration to create new components.
If your alignment is creating too many components than it might indicate image overlap problems or bad RC alignment settings.
I’d recommend to take a look at this information: https://support.capturingreality.com/hc/en-us/articles/115001528211-Taking-pictures-for-photogrammetry
Hi Artur,
this topic has been discussed a lot in the past. The more random your results are, the closer your image set is to breaking up, meaning it is not ideal.
Michal,
I am reasonably certain that there is no info for alignment stored in the cache, only in the components. Or rather the info stored in the cache is only valid for each component individually and will not be used for others. In my experience, if you delete all components, it is a fresh start from scratch. But I might be wrong. Where did you get this info?
Hi Michal and Götz
Much obliged for such fast replies.
@anonymous_user_4eb56474 : the process actually creates too little components and thus portrayes an incomplete point cloud. And then through what seems as a completely random process of a few different Alignments i get all pictures aligned and the model looks amazing.
I will check those tutorials again for sure but let me ask directly because my usecase might be different to most people asking here : Im working for a gaming studio and we are trying to scan a huge construction site. We dont have optimal equipment so we have to work with what we have. Scanning it all is out of the question so we scan individual objects and assemble scene elsewhere. It works great apart from floors. No objects just ground with detail. I cant seem to find a way to scan blocks of 10X10m the usual way as if I am scanning an object. I take 100 images on average at varying distances completing loops as the tutorials suggsst but RC always gets lost somehow. So im experimenting with control/tie points to manually tell it which parts of the scene are connected…
Kind regards
Hi Artur,
some screenshots (images and model with camera distribution) might help for a better evaluation.
Depending on your technique (drone?), resolution and intended LOD, 100 images for 10x10 m might not be enough…
Oh and yes, having first one component with a fraction of the images and then including more after another alignment is totally normal and absolutely a recommended workflow. It means that your image set is not perfect but RC can still get it together somehow… 
@Götz Yes you are right.
Just to clarify: RC cache is storing only feature detectors after the alignment process. So If you clear RC cache but not previous components, the next alignment iteration will take existing components into calculation to rematch the input data differently. In short nothing will change in terms of alignment process if you delete the RC cache or not.
Ah ok, thanks Michal!
OffTopic:
You are not Michal from RC, right?
@Götz Yes I am a Technical Support at Capturing Reality.
Ah, ok. That explains the activity! 
Wow I really appreciate the attention and help you guys emit around here…really impressed and thankful
So anyhow, to not waste your time with theoretical questions, and since Im already working at the site, I went ahead and made a reference shot sequence of a 7X10m ground square. The images can be seen here (hopefully the google photos link is working) :
https://photos.app.goo.gl/Upfa8i5GWrJnC26d2
When I import them in RC on my laptop just now, it took some 3-5 Alignments to get the location of the components right but now the orientation is switched 90 degrees. The actual “floor tiles” aren’t aligned properly as well, so Im trying to use this example to a) learn from you guys how to take pictures more appropriately for my situation and b) to use tie/control points and correct the mistakes in RC estimates - if it makes sense for the case, of course.
In this picture I circled which components are misplaced and direction which they should be in yellow (it was not the order in which I took the photos) : https://imgur.com/a/GPYOKvE
Best regards
Hmm, weird.
Could be the flat angle.
What is your goal here to capture ? I mean you want to capture whole building interior or not ? If so why do you capture such 10x10m flat tiles only ?
I’d not recommend capturing these tiles as you did, especially the X pattern that you made with angled shots might not be optimal for RC since you don’t use CPs (GCPs) and the connecting shots between those 2 shot lines and the loop is missing. The loop around is OK but make sure you have connecting shots that go gradually from the loop to the X pattern and back. But overall I wouldn’t do the X pattern at all. Instead you might want to do zigzag topdown pattern as if from drone mission with connecting angled shots to the loop around. 80% frontal and 80% lateral overlap is ideal. But frankly without a drone and GPS signal you won’t get such pattern indoors unless you calculate it and measure before and fly it manually over inhand. Understand what I mean ?
The second option is to use CPs. Print some markers and place it in each corner of the tile and one in the middle. Make sure each CP is visible at least on 3 images. This should help and omit the need for zigzag topdown pattern.
Michal
Thanks so much for the extended response!!
My goal isnt the complete interior but just the ground. I made the tiles as example for our discussion so that you guys understood my question better, on an example.
For doing ground only do you have a more optimal way besides doing smaller tiles? I have even less success trying ti scan bigger areas at once.
Will definitely try your advices regarding positioning and methodology of taking photos. As well as printing and placing markers…