Agisoft Metashape
Agisoft Metashape => General => Topic started by: ppkong on July 02, 2014, 06:15:25 AM
-
I use oblique aerial photography to generate three dimensional model of building, specify the parameters of the camera and imported the POS data, without ground control points. Generate a model of the building as follows:
(http://)
But it always felt good models that aren't GoogleEarth effect, the edges of the building not obvious enough.
I think Google has used a similar method to generate a three dimensional model of the following figure is GoogleEarth models.
So I want to know how can improve the precision of the dense point in the PhotoScan and generated mesh better?
-
Which mesh generation method are you using? It looks like you are using the 'Heightfield' setting, and this will lead to less defined vertical surfaces. You could try 'Arbitrary' (but it will take a lot longer and more memory is needed).
Please see this thread with a similar problem:
http://www.agisoft.ru/forum/index.php?topic=2505.0 (http://www.agisoft.ru/forum/index.php?topic=2505.0)
-
I just used this option and I used the following settings:
1.Align Photo:
Accuracy=high;Pair preselection=Ground Control; Point limit=40000;
2.Build Dense Cloud:
Quality=medium;depth filtering=aggeressive;
3.Build Mesh:
Surface type=Arbitrary;Source Data= Dense Cloud;Polygon Count=high;
4.Build Texture:
Mapping mode=Generic;Blending Mode=Mosaic;Texture size=8192*2;
-
Try
2- HIGH - Moderate ( ULTRA + Moderate on part of the project )
Google use commercially not aviable technology, very pricey and specialised for their needs + captured mostly with 120 +50 Mpix LArge cams setup....
-
Google use commercially not aviable technology, very pricey and specialised for their needs + captured mostly with 120 +50 Mpix LArge cams setup....
120 cameras? Do you have more information on this, I would love to see this setup :)
-
Have dataset of Paris, 120Mpix UltracamX + 50Mpix Haseblat - Oblique imagery. https://www.dropbox.com/s/ecvppyt0v63zbhg/Screenshot%202014-07-02%2011.39.35.png
-
Have dataset of Paris, 120Mpix UltracamX + 50Mpix Haseblat - Oblique imagery. https://www.dropbox.com/s/ecvppyt0v63zbhg/Screenshot%202014-07-02%2011.39.35.png
Ah, a 120MP camera, I thought you meant 120 times a 50MP camera. I had a hard time figuring out how that would fit in a plane.
But still impressive (I'm surprised by the neat alignment of the aerial pictures)
-
@Wishgranter: where you got this dataset from?!
is it possible to shere, just to play around... :-P
-
Hi, cannot say how im get it - officially have right to "use it" and test but cannot share..... you can see result with help of Google EARTH, just enable 3D buildings and zoom around
https://www.google.com/maps/place/48%C2%B051'59.2%22N+2%C2%B021'14.1%22E/@48.8670174,2.3534072,2035m/data=!3m1!1e3!4m2!3m1!1s0x0:0x0
-
Try
2- HIGH - Moderate ( ULTRA + Moderate on part of the project )
I tried to modify this parameter to high, but the increase in processing time by a factor of 4 times, really depressing, do you have a method of optimizing?
-
Try
2- HIGH - Moderate ( ULTRA + Moderate on part of the project )
I tried to modify this parameter to high, but the increase in processing time by a factor of 4 times, really depressing, do you have a method of optimizing?
Photoscan downscales your source images based on the quality setting. At Medium it will use the images at 25% of the original size, and at High it downscales to 50% of the original size. So there are 4x as many pixels to go through.
At Ultra it will use the images at full resolution (but your images need to be perfectly sharp for this, otherwise you will just get more noise)
The only way to optimize is to use less cameras, so if you have a lot of overlap between your images you can try to disable some cameras manually. Using 2 GPUs can also speed up things a bit (maybe a 60% speed increase if you run your scans on High. It does not speed up linearly with the number of GPUs because the Dense Cloud reconstruction also has a phase when it only uses the CPU).