Agisoft Metashape
Agisoft Metashape => General => Topic started by: Smallpoly on November 08, 2017, 01:39:01 AM
-
I'm having some fun testing out these new features and seeing how they compare to the old methods. I thought I'd post some side by side shot of the models I'm testing with as things come out. Feel free to post yours too.
First up is "Visually Consistent Mesh Generation", found in Tools -> Preferences -> Advanced -> Use Visibility Consistent Mesh Generation Method (Experimental).
I'm testing with a data set I have from a kitschy house model I picked up from a thrift store. Some of the pics in the set weren't as clean or as in focus on the right parts as they could have been if I were to do it again, and the lighting was still off a little, but it still worked out reasonably well. Here's a crop of one of the pics.
(click for full size)
(https://i.imgur.com/rCSHn9D.jpg)
Here's a comparison where I've generated everything at the maximum quality I could get the set to produce. On the left has alignment and dense cloud maxed out and the model generated at 4 million to get as much detail in as it would support. On the right is using the new method at max quality.
The really interesting thing about the new method is that it doesn't actually need a dense cloud to work. This is also without using the new mesh refinement option since I haven't gone into testing that one yet. The visibility consistent mesh did a way better job of getting the little toothpick poke marks in the roof and the right shapes of the bricks and other little indents.
(click for full size)
(https://i.imgur.com/ZLbxVAP.jpg)
Live Model:
https://sketchfab.com/models/c1fef8c991d4459d8e0ab68499f17665
-
Here's a new one that does a better job of showing what the new mesh generation method gets you - it works better with thin and intricate objects, such as plants. Comparing these two shots from the same data set, the model on the right does a much better job of following along the "leafs" of this office plant with half the polycount. This was handheld so the source images are a little rough.
(https://i.imgur.com/6OM5qSX.jpg)
-
Thanks for the comparison!
I didn't even know that function existed, despite toying around with the beta for a bit.
-
where did you download 1.4.0? I can only find 1.3.4 on the donwload page
http://www.agisoft.com/downloads/installer/
-
At the top of this forum.
-
Grin - the 1.4.0 pre-release is at the top of this thread, stickied to the top of General:
http://www.agisoft.com/forum/index.php?topic=7730.0
-
Next up, here's a comparison between Mosaic texturing with and without the new ghosting filter option. Average is also added for comparison. Filtered mosaic has a very clean result while staying very sharp. Average is also fairly clean, but much softer of a result.
The test object I used here was the ceiling of an interior I shot a while back that had a bunch of hanging wires and other things. In that project I ended up having to do a lot of cleanup in Photoshop with healing brush and other tools. This would have had a nicer result and saved a lot of manual work as well.
Unfiltered Mosaic (click for full resolution)
(https://i.imgur.com/yXruTaU.jpg)
Filtered Mosaic (click for full resolution)
(https://i.imgur.com/PgBn7ox.jpg)
Average (click for full resolution)
(https://i.imgur.com/5ACIZoS.jpg)
Animated Comparison (click for full resolution)
(https://i.imgur.com/3EXas10.gif)
-
Wondering about Tie Points new creation...
Agisoft 1.3.2 creates from a model in High mode 1.800.000 Tie Points
Agisoft 1.4.0 creates in High Mode only 470.000 Tie Points
How do I manage this differences?
-
Wondering about Tie Points new creation...
Agisoft 1.3.2 creates from a model in High mode 1.800.000 Tie Points
Agisoft 1.4.0 creates in High Mode only 470.000 Tie Points
How do I manage this differences?
Do you have processing logs from both runs or at least Chunk Info information with the parameters used?
-
if you tell me where I can find the logs and how I can save them with the laptop in demo Mode (due to this stupid licence concept) I can try to send you the logs.
-
Hello GrinGEO,
If you have "write log to file" option enabled in the Preferences window - then the processing log is available in the specified location. Otherwise you can copy the processing log from the Console pane (in PhotoScan window) and save it as text file.
-
Unfortunately Agisoft crashed and I lost my log.... so Im sending you the report informations....
Processed on Agisoft 1.4.0 b5310
Points 463,402 of 496,538 RMS reprojection error 0.422863 (1.11721 pix) Max reprojection error 6.33695 (62.3609 pix) Mean key point size 3.14836 pix Effective overlap 3.97304 Alignment parameters Accuracy High Generic preselection Yes Reference preselection Yes Key point limit 40,000 Tie point limit 1,000 Constrain features by mask Yes Adaptive camera model fitting Yes Matching time 35 minutes 18 seconds Alignment time 42 minutes 42 seconds
Processed on Agisoft 1.3.2 b5310
Point Cloud Points 1,864,722 of 1,921,253 RMS reprojection error 1.55935 (4.00157 pix) Max reprojection error 151.041 (289.275 pix) Mean key point size 3.44991 pix Effective overlap 3.57946 Alignment parameters Accuracy High Generic preselection Yes Reference preselection Yes Key point limit 40,000 Tie point limit 4,000 Constrain features by mask Yes Adaptive camera model fitting Yes Matching time 46 minutes 54 seconds Alignment time 42 minutes 42 seconds
-
Hello GrinGEO.
According to the parameters in the version 1.4.0 the tie point limit has been set to 1000, whereas in the version 1.3 the value of 4000 has been used. So it gives corresponding difference - approximately in four times, for the total number of matching points in the cloud.
In case you have any information regarding the crash - it would be helpful to find out the reason of the issue.
-
Ok I understand. Is in 1.4 standard only 1000 Tie Point Limit or could this be an error that I made? Just to know to avoid this next time...
Thanks for helping
-
Hello GrinGEO,
The default tie point limit is 4000 since several previous versions.
The recently used values are not reset when you install the newer version, so 1000 could be used previously by the same edition in the previous run.
-
Thanks for posting Smallpoly. I too missed that one. Looking a little deeper now.
[Update] Here's my mesh comparison:
(http://www.digital-mapping.net/forums/Photoscan/2017/PS140_mesh.jpg)
-
The new mesh after Refine Mesh...
(http://www.digital-mapping.net/forums/Photoscan/2017/PS140_refine.jpg)
-
For now, it seems that this feature produces interesting meshes for isolated objects but not so much for landscape features, using images taken from UAV's. Building a final DEM shows how huge the difference is.
Cheers
-
Following this with interest. Having poorer quality DEMs for UAV work really concerns me.
-
Hi Guys i did some experiments use the new mesh generation+ volumetric mask + refine mesh
Looks very promising feature.
Best
Mohamed