Only one side photos can be aligned

[zhycever]

I use version 1.0.3 pro to rebuild a cactus, 42 photos surround cactus but only 25 are aligned, aligned photo mainly on one side. I tried many times, but result changes a little.

[Alexey Pasumansky]

Hello zhycever,

Could you show a few thumbnails of the photos from the set? Please also specify if the object was static and camera has been moving around it or you have been rotating the object.

[James]

Sometimes i find when only half of images are aligned, it is not because half the images are bad but because photoscan finds two subsets of images that can be aligned, only not with each other. Since it cant align the subsets to each other it just shows one or the other and leaves the other 'not aligned'

to see if this is the case, duplicate the chunk, then reset alignment for aligned images in the duplicate chunk, now select the images in the duplicate chunk that were not aligned previously and right click -> align selected images. You may find that now the other half of the images are aligned in this duplicate chunk.

if you are lucky you may be able to align the two chunks together using the align chunks workflow process, or you may have to take some more photos at the joint between the chunks where the overlap is not sufficient for whatever reason. if you have pro version then you may be able to align using  markers.

[Wishgranter]

try capture more images, say 60... later can DISABLE what will be used for reconstruction...... and  capure from say 2 levels of height ( or even more 4-7 )

[Marcel]

Also, can you show us some photos at full resolution?

[zhycever]

cactus was static and i hold camera moving around. I shot picture at 2 different level  for its simple shape. i just take another set photos and try again. this time i take 42 pictures at 2 level and only 10 are aligned, that's pretty bad. i use a ADC camera, is this a problem? or should i change to 0.9 version?

[Wishgranter]

Just a bit curious, why is the cactus lighted with pink color just from this image i se ethat it have problem to find enough points, because the pink color "flatten" the image.... can try with normla lighting ??

[James]

Is this a multispectral camera? I have to admit i know nothing about those, or using them in photoscan although i understand that people have done.

However another problem is that the background is very flat/texureless.

If you could place it on a carpet/newspaper/magazine then it would help alignment, and will result in a few more valid matches on the cactus surface.

It also looks like more light would help, but i'm guessing you are in a limited environment.

An easier version of my previous suggestion is to import your photos in chunks, about 10 at a time, with 2 or 3 photos common to successive chunks, see if each chunk of 10 photos will align individually and then try to align the chunks. It may also help you see where the problem is.

I often have to work with awful datasets and you have to give it a nudge sometimes.

[hengefjes]

Found this in a faq about ADC-cameras: "The ADC is limited to capturing the spectrum from 0.52 microns (green light) to the near-infrared, no blue wavelengths are captured. A "regular" digital camera is designed to see only the visible spectrum but cannot capture near-infrared wavelengths."

Could it be a problem with repeating patterns? Maybe adding some random leaves around the bowl would help.

I guess there is a reason why you are doing this with the adc-camera, but in this case I think your model would be better with a "normal" one.

[frank.stremke]

hello
looking at your picture besides the cactus there is nothing for a goof alligning process
i would recomend puting the thing on a surface with some very good texture like a non glosy newspaper or so and try to get as much diverse surfaces as posible maybe also in the background
a single colored cactus on a dark featureless tabel in a pot of the same type is not at all good for allignment.
you can cut of the newspapers after allignment from the modell or by using masks (or both :-))
frank