still separate components

Hi Guys

I´m trying to merge two components but with no success,
i tried with control points but i think i´m doing something wrong, i dont know if masking would be ok because each component is correct,
is a turntable and have two cameras:

https://drive.google.com/drive/folders/1I1oXpQk8UEIKBWheBx3gAiekT8FDAwtp?usp=sharing

any help is appreciated enough
thanks so much

hello Enrique,

I recommend you to read these:

https://support.capturingreality.com/hc/en-us/articles/115001485012-How-to-put-together-more-components-

https://support.capturingreality.com/hc/en-us/articles/115001569011-Working-with-Components-Merging-components

Hi Enrique,

similar problems as in your turntable thread: extremely difficult materials everywhere, too small coverage, blurry images.

If you got all images in 2 components, I think that’s quite good for this image set I would say!  :slight_smile:

Try to set the preselector to 20k - this helped me once…

Thanks so much guys fot your advices, i will stil practicig here is another test

regards,

Well this one was much more successful!  :slight_smile:

Another hint for that one: There is also too many images for a good, as in smooth, result. The reason for that is that software usually uses a neighboring image pair to calculate depth information. Iif the cameras are too close, there might be more errors since the basis of the calculations (distance) is way too small in relation to the distance of the model. It’s pure geometry. So I think you might get a better result if you left out at least 2 of 3 of your shots. Rather do 2 more circles in between and one on top wouldn’t do much harm either, I think.

thanks so much Götz Echtenacher

10-15° between images is a good value. You should take between 12 and 36 photos per revolution depending on the objects complexity.

More will actually hurt the quality of the reconstruction as Götz said.

Take extra shots where the initial reconstrucion shows missing detail as needed.

thank you ShadowTail i will keep it in mind!

Enrique, no problem.

I’m just sorry that I can’t say anything better.

But that will probably change soon, right?   :slight_smile:

 

Götz Echtenacher said:

“… software usually uses a neighboring image pair to calculate depth information. Iif the cameras are too close, there might be more errors”

By neighbouring do you mean with adjacent filenames i.e. photos taken one after the other - or do you mean simply neighbouring in space, even if taken later as an out of sequence fill-in photo?

How important is it to take photos in a continuous series - or does RC routinely check through the whole photo set to find neighbours in space?

For example, taking an upper row along a wall then coming back with a lower row, properly overlapped vertically same as horizontally, finishing back at the first position of the top row, to complete the circuit? Any problems with that? Surely the vertical spacings should be same vertically as horizontally - a square grid of cameras, or maybe 3x2 (or 4x3) rectangular grid because of photo format?

Apart from that, you’re saying only one photo within 5o or 10o of each other and actually remove any nearer-identicals/duplicates? If so that’s a good tip, easily done. And I’d say, according to Help, no gaps wider than 30o from at least one other photo (if that’s advised for turntable increments, must make equal sense when walking parallel along a wall).

I don’t know how they are chosen, I think that’s the “matching” step though, so it’s likely that they are calculated somehow.

I’ve read before  that taking images in a row would be better, but not how or why (as usual). I could imagine that it is mainly for speeding up calculations since RC might check first if the next image fits to the current one and if not needs to search all others to find a match. But again, that’s mostly speculation…

The software does compare each picture with every other picture to find the ones where features match up well enough to be able to triangulate them and the camera positions with acceptable error/accuracy.

Great, so I’ll take it for now that having photo filenames in strict sequence e.g. along a wall may help a bit with ‘matching’ speed but isn’t significant for end-result.

About avoiding photos which are too close (less than 5o or 10o?) to each other in vector. Been trying looking in 3D for those and it’s not straightforward.

What about 2 photos like that but one of them taken at greater distance from the subject? That significantly alters the geometry, one to the other. Enough?

Because when photographing at an internal angle between say end wall and side wall of a room, I step along one wall nearly to the corner. Then I need to start the other wall also from the corner but turned 90o. But in between, I have to make shots in to the corner at 30o and 60o to link the two runs - yes? In doing this it’s hard to avoid creating effective panorama shots paired with the penultimate photos (almost from same viewpoint but turned 30o) of the two straight runs. Backing out or closing in is a way to ‘scatter’ the cameras. Any good?

In fact if the last shot along each wall is turned 15o towards the corner, those shots are only 60o not 90o apart, so only need one linking shot at 45o, not two at 30o and 60o.

With such systematic coverage, am actually getting half-good results along smooth-plaster walls that previously defeated me. Seems that fantastic pixel-density (to pick out paint brush-strokes on the plaster) isn’t going to be necessary - good overlap and tripod-mounted sharpness may be enough. Next to try increasing ISO above 100 - at ISO1600 hand-held may be OK.

For rooms the recommended strategy is to actually go round the room with your back to the wall, taking pictures of the opposite wall.

That way you get the required overlap so RC can make the proper connections.

Do some closer up photos as well for all the fine details.

But you want some form of transition between the far away and close shots. Too much of a difference and RC won’t be able to align them.

‘back to the wall’ depends on room width and other practical considerations. Not right in a long thin room for instance. Much of the far wall may be occluded in a long shot.

Recommended range would be more useful, depending on pixel density appropriate even for the big-picture shots. Proper overlap can be done at any range.

Elsewhere I’ve seen the internal corners transitions carefully done as I described.