How to Identify misaligned images?

Development Asset Creation
,
1

Hi,
This is a problem I have run into a few times, particularly when using a turntable but I haven’t yet found a solution.
Sometimes I find errors in the model that look like a small area of heavy noise, but it looks almost ‘projected’ onto the model from a single view. I assume this is caused by a camera misalignment? If so, how can I identify this image out of the hundreds or thousands of images I have in the chunk?
Thanks

Jules

2

Hi julesmalcomson

Can you provide us with some screenshots ?

3




These were rendered at normal quality.
When I rendered the same model on high these artefacts were gone, which suggests I was wrong in thinking it was a misaligned image.
Any ideas what causes this? The patterns didn’t seem to correspond to the model chunks which is what led me to think it was a bad camera causing the problem.
All settings were on default for this.
Thanks

4

Any ideas what causes this? I know this can generally be avoided with more photos but I’m still curious to know what causes this, it will add to my understanding of how RC works

5

Not a experienced CR user, but if i have similar issue in other tools, this usually happen if such place have more detailed shots than surrounding areas. For example area around shoot far from object, but this area shoot from small distance. And this is probably depend on sharpness and details of this images. Also this areas have more dense polygon count.

6

Hi julesmalcomson
If all images are SHARP, then it is a problem of misalignment of a few images, as RC is calculating and weighing depth maps, so if you get this on NORMAL, then it can be suppressed on HIGH as there is 4 times more data for the final mesh calculation and they can overweigh the trouble-making images-depth maps.
How to identify misalignment ? It depends on a scene, but sometimes it can be really problematic ( depending on too many factors ) - in this case is good to do the alignment in parts. Is it a big dataset or just small one ?

7

It’s only about 300 images. The object was very glossy so I had to ‘texture’ it using developer spray. The model features down to to sup pixel level and every image was pin sharp at 36mp, but the object was definitely not ideal for photogrammetry.
Do you mean that I should run alignment in chunks of around 5, and then use control points to align manually align the components?
Is this to help me identify the bad images? or does it just help RC if the cameras are aligned in smaller groups?
Thanks

Jules

8

Hi Julesmalcomson
So the culprit of the missalignments is found :smiley:

Is it possible for you to put newspaper under the scanned subject ? Powder is a good solution, but the drawback is that it can create repetitive patterns and so you can get to missalignments…