1.7.0. Results of reworked depth maps generation algorithm

[Arie]

Hi Agisoft team,
congratulations on the 1.7.0 release. I've been doing some comparisions between 1.6.5 and 1.7.0 and I thought I would share the results.

Agisoft 1.6.5: 12h 8m (Depth maps 9h 57m, Meshing 2h 11m)
Agisoft 1.7.0: 7h 14m (Depth maps 5h 9m, Meshing 2h 5m)
Competitor: 5h 11m (Depth maps 3h 5m, Meshing 2h 6m)

In general the improvements are quite impressive: Significantly shorter processing times with more detailed results. In addition, the stepping effect visible in some of the 1.6.5 images vanished.
But some areas of the 1.7.0 model show artefacts that were not present in 1.6.5 (last image). Any fixes in sight?

Also I am very impressed on how much Metashape has caught up to their main competitor. Keep it going! <3
Cheers!

[Mak11]

Arie,

Thanks for posting this info. Can you share you system specs and also project info (number of images/resolution etc).
I'm also encountering the same issue as seen in the last image. I'm wondering if it's related to the interpolation. Having it on disabled or Auto(enabled) or extrapolated doesn't seem to make as much difference compared to 1.6.5.
As a matter of fact I'm now restarting to have quite a bit of unwanted geometry generated inside the meshes (or outside) as was the case in 1.6 (pre-release I think).

Mak

[Arie]

Hi Mak,
i've made the same observation regarding the additional geometry on the inside.
The image set is made up of 258 61mpx images shot with a 90mm lens on full-frame.
System is a Threadripper 3960X, 256GB 3000mhz DDR4, 1x 1080ti, 2x Samsung Evo Plus NVMe 1TB Raid0, Win 10 Pro.
Anything else?
Cheers!

[dpitman]

Also I am very impressed on how much Metashape has caught up to their main competitor.

Which is?

[Arie]

Let's just say there aren't many other programs out there that are quite a bit faster than Metashape... I don't think we should elaborate on other software here.

[dpitman]

I'm not trying to cause trouble; I'm really curious.  I have only used Metashape for topography and not model creation like your example. So, perhaps the mystery software is not suited for topography or maybe it is. That is why I am asking.

I've used Correlator 3D, and it is a faster processor than Metashape.  But it is more focused on topography rather than a multi-focused product like MS and so I doubt that is what you are referring to.  I've tried Reality Capture, and it had some issues and an email to support went un-answered, so that wasn't impressive.

I am a mod on the Drone Deploy sponsored forum and we point folks to alternative solutions all the time if it is something that will benefit them.  Obviously, MS can stand on it's own value.  I guess I'm not understanding the apprehension to discuss the marketplace, but what ever, I guess.

[Mak11]

Diptman,

The other software is Reality Capture.

Arie,
Thanks for the specs & info.
With a newer GPU (RTX2xxx/3xxx or AMD RadeonVII/RX6XXX) your Depthmaps computation time in MS 1.7 will probably be close to RC.

Mak

[dpitman]

Thanks Mak

[Arie]

Mak,
this build has a free Pcie slot that's just waiting for the 3080ti to be released

[jedfrechette]

Let's just say there aren't many other programs out there that are quite a bit faster than Metashape... I don't think we should elaborate on other software here.

Why shouldn't we be able to mention other software? This whole poisonous idea that you shouldn't talk about other software originated on the RealityCapture forums where the mods put in place draconian rules to that effect. Nothing exists in a vacuum and pretending that it does is foolish. All of the photogrammetry tools have improved significantly in the last few years and I think that is in no small part due to the competition between them and the ability to see what does and doesn't work well in the various packages.

One thing I really appreciate about Agisoft is the way they interact with their users, particularly on this forum. They are very receptive to feedback and constructive criticism, which ultimately makes their product better. That is distinctly different than the RealityCapture forums where users get banned for voicing criticism, which I think is incredibly shortsighted on their part. In the long run the best software is always developed based on user feedback and if the developers aren't capable of collecting and listening to that feedback their software will not improve as efficiently as the software of their competitors who do.

[Paulo]

Well put Jed! completely agree. Censuring constructive criticism will never bring meaningful improvement.

[Arie]

Why shouldn't we be able to mention other software?

While I am definitely on your side here, it just seems like the wrong place to discuss competing software packages on the forum of one of the competing solutions. I would much rather prefer such discussion to take place on more neutral grounds, similiar to what is being done on laserscanningforum.com.

One thing I really appreciate about Agisoft is the way they interact with their users, particularly on this forum. They are very receptive to feedback and constructive criticism, which ultimately makes their product better.

I completely agree with you. That's why I also "dared" to add the processing time of a competitor to give some feedback on the improvements but also to show that there is still room for further optimizations.

[Mak11]

Alexey,

Using the Banana sample project you can clearly see that the new algorithm is once again generating unwanted geo all over the place.

Mak

 

[Mak11]

Arie,
Can you share the data-set?
I would like to run it on a more powerful GPU to see how 1.7 scales.

Mak

[Arie]

I just released a similar image set under CC-BY-NC 4.0 for testing and comparing to other software. I have some high-res 3D-scans of that model, so it's great for comparison.

JPEG (2x 2GB) incl. masks
https://drive.google.com/drive/folders/1qe1jZg6iZ_yeZ865VL83KXpsV5Xifih1?usp=sharing
RAW (6x 2GB) incl. masks
https://drive.google.com/drive/folders/1EJQpSnqHu3XSb7R4FmJzuRB5O8ixXwka?usp=sharing