Meshing a 10 story building
Answered
A simple post of my experience on this test:
Last week I took pictures of a 10 story building from the building opposite.
I took 190, 20mp with my sony rx100M3. I took images from the ground floor, 6th and 9th floors.
Firstly I found I couldn't import the raw files so I had to use the .jpgs.
CR took about 48hrs to mesh the images in high detail, eating about 63 Gigs of memory on my MSI GT70 24gig laptop.
The photos were mainly of the front face of the building with some spill on two other sides.
I found the resulting mesh had a lot of trouble distinguishing between positional information and the texture on the building. In fact part of the building was in stone and part was very smooth matt, the smooth part resulted in a very poor mesh, worse than the stone part.
The resulting mesh of the building was also sheered in two directions.
After this test I went back to my laser scans which produce much much cleaner results ;)
Last week I took pictures of a 10 story building from the building opposite.
I took 190, 20mp with my sony rx100M3. I took images from the ground floor, 6th and 9th floors.
Firstly I found I couldn't import the raw files so I had to use the .jpgs.
CR took about 48hrs to mesh the images in high detail, eating about 63 Gigs of memory on my MSI GT70 24gig laptop.
The photos were mainly of the front face of the building with some spill on two other sides.
I found the resulting mesh had a lot of trouble distinguishing between positional information and the texture on the building. In fact part of the building was in stone and part was very smooth matt, the smooth part resulted in a very poor mesh, worse than the stone part.
The resulting mesh of the building was also sheered in two directions.
After this test I went back to my laser scans which produce much much cleaner results ;)
-
BuildingTest_01.pngBuildingTest_01.pngBuildingTest_01.png
To hazard a guess on my results here I think I know why my results are this way.
The images I submitted for this mesh vary from 9th floor shots from the building opposite to shots from the ground.
Of course the shots from the ground will have much less detail and are probably getting mixed up with the CU shots I took from the 9th floor.
Similarly the shots I took from the 9th floor looking down to the bottom of the building are of course much less detailed the further away the building is.
Also, my shooting angles were restricted and due to the subject often quite extreme.
If RC doesn't differentiate between hi-res and hi-detail sections of images and lesser detailed sections of images then of course this will result in a soup of varying quality which doesn't line up so well.
Are any of my assumptions correct ?
It looks like I need to do another test to better simulate the camera positions of say a "light stage" shoot.
Using the same focal and keeping the same distance to the subject using loops of images around said subject at different heights. This technique probably delivers the best results.-
56BuildingTest_01.pnghttps://www.capturingreality.com/forum/download/file.php?id=562097283imageFile pngBuildingTest_01.png
-
57BuildingTest_02.pnghttps://www.capturingreality.com/forum/download/file.php?id=572097283imageFile pngBuildingTest_02.png
-
58BuildingTest_03.pnghttps://www.capturingreality.com/forum/download/file.php?id=582097283imageFile pngBuildingTest_03.png
-
-
Hi Peter
sorry for late reply, have plenty of work and just checked if have replied to everything...
IN short if have say imgs of subject taken in 10m and looking on small detail say 1 cm
then if get closeups from say 7 -5 - 2 Meters away we will "add" the details from the closeups
just from the results it look like have not "enough" images-overlap there...
can post a screenshot so can see the CAMERA positions used ??
Please sign in to leave a comment.
Comments
4 comments