Suggestions to improve model quality needed.

[jonc]

Alexey took a look and my files and came back with this. 


It just goes to show how important experience is, his result is 10x better than mine!

I've taken a few more close up photos of the lace and welt areas and I'm going to have another go using what I've learnt.

[Alexey Pasumansky]

Just to specify our steps with this dataset:

- first of all the markers were removed,
- then we realigned the image set,
- then removed only a few hundreds of the outliers and points with the highest reprojection error,
- then optimized the alignment,
- then generated dense cloud and cleaned it a bit, removing unnecessary background that wasn't masked on the images,
- then generated mesh.

So actually nothing special. I assume that the issues could be related to the use of the markers leading to the affected alignment and typical for that artifacts on the mesh (like corroded surface). Also I believe that image quality improving (like using lower ISO) would result in more smooth and accurate surface reconstruction. And you can use the image frame space more effectively.

[Photogrammetryfacts]

I know this might be a bit of a strech, but would you be willing to share your data set? I'd like to try and replicate the results from Alexey to see if my workflow will get the same results. That way I could know whether or not my workflow needs improvement.

[jonc]

As I was having problems getting an overall alignment I added markers. As that improved the result I kept on and added masks for all the images, i think this is where the problem was.

As Alexey says, if I had just masked the images first and let the software do it's job I would not have had the alignment issues at all. I now know this for the future.

I've been experimenting some more and getting much superior results.


Photogrammetryfacts PM me

Just to specify our steps with this dataset:

- first of all the markers were removed,
- then we realigned the image set,
- then removed only a few hundreds of the outliers and points with the highest reprojection error,
- then optimized the alignment,
- then generated dense cloud and cleaned it a bit, removing unnecessary background that wasn't masked on the images,
- then generated mesh.

So actually nothing special. I assume that the issues could be related to the use of the markers leading to the affected alignment and typical for that artifacts on the mesh (like corroded surface). Also I believe that image quality improving (like using lower ISO) would result in more smooth and accurate surface reconstruction. And you can use the image frame space more effectively.