Jump to content

Reality Capture - New photogrammetry software package.


leszekp

Recommended Posts

Reality Capture (RC) is from the team that developed CMPMVS, an excellent multi-view reconstruction program (still available in an earlier version for non-commercial use - works well with VisualSfM as the front-end for photo alignment). Available as a free beta through 1/31/16 at their website. Played with RC a bit - interface takes some getting used to, help file isn't quite complete yet. Models I've generated are comparable to or slightly inferior to those of Photoscan, but this may improve as I climb the learning curve. Seems as though Photoscan is better at photo alignment than RC. Works with laser scans as well. One big thing in RC's favor is that it's fast, at least several times faster than Photoscan. Says it supports distributed computing, but no info yet on how to get that to work. 

 

Edit: Requires an nVidia graphics card wtih CUDA capability.

Link to comment
Share on other sites

I have been testing Reality Capture just a bit. Unfortunately I can only currently test it using Windows 10 in a Parallels virtual environment so I have not tested anything beyond alignment. However, learning how good the alignment and subsequent camera calibration is really my biggest curiosity/concern.

 

Everyone seems to think that RC is faster for alignment but I am not so certain. I believe that by default the alignment is done using down sampled images. At least half the original size, perhaps even more. If you do an alignment using PhotoScan at Medium or Low I believe you will see nearly the same speed increase. With that speed of course there is a compromise. The camera calibration will never be as good as is possible on High. I of course am a stickler for the highest quality camera calibration as that is the single biggest source of error for a photogrammetry solution. Meshing of course will always be faster if CUDA is available and RC may very well be faster at meshing than PhotoScan and others but I cannot test that currently. I would be very interested in any testing that others might do.

 

That is not to say that RC is not good. I think it is quite good and fast. I am not sure it is better than what is already available but it is too early to tell. It will be interesting once they announce pricing. Of course PhotoScan is available and runs well on Macs which is a big deal for me. It will also be interesting to see if RC becomes available on something other than Windows and without the CUDA requirement.

 

I hope to do more testing before the end of the trial period but there never seems to be enough time. The interface is a bit different. A little more black box than I would like. It seems a little difficult to figure out what is really going on and how to make sure of the quality of the results.

 

If anyone else does some testing, please send some feedback.

 

Tom

Link to comment
Share on other sites

Well, that was entertaining. Ran a large set of photos (about 2000), took about 30 hours for full processing. got a decent looking model, saved the project. When I reloaded it, the mesh, texture and the photo alignments were gone, even though there's a large set of data files associated with the project. Fortunately, I had exported a mesh and an orthophoto, but I'm not going to waste any more time on this program again until it's in better shape.

Link to comment
Share on other sites

  • 2 weeks later...

... When I reloaded it, the mesh, texture and the photo alignments were gone, even though there's a large set of data files associated with the project.

 

I had this when after my first model, but realised that it if not all of the images align into a single model you will get multiple components. By default after the alignment it shows you the component with the highest number of aligned images. When you open a project it shows you the first component, which may only contain a small number of images and thus not show much.

 

I've been using it for a couple of weeks now. I stopped a Photoscan project 5/6 of the way through creating a dense cloud 9 days after starting the initial alignment (~800 20mp images). I had a mesh using RC within 12 hours.  Without much documentation it can be pretty quirky to use but it is definitely worth the effort.  It splits tasks up into manageable chunks for your hardware extending the range of possibilities and alignment is very fast.  Detected points in images are cached for the session (even if you start a new project) so tweaking and realigning doesn't have to wait for point detection each time.

 

There is now a Facebook group for sharing experiences and getting help: https://www.facebook.com/groups/CapturingRealityArena/

 

For me, this is potentially a game changer. Increasing productivity 7-10X and coping with at least 2X the number of images for a project on the same hardware. Definitely one to watch.

Link to comment
Share on other sites

Thank you! I guess this makes sense, but the lack of documentation doesn't make this intuitively obvious.

 

So I have multiple components, and I'd like to merge them using manually-added image control points. Is this possible with the current version? The help file says yes, but I'll be damned if I can figure it out how to add a control point.

Link to comment
Share on other sites

https://www.youtube.com/watch?v=d8naLEtLqDY

 

Each pane can be set to 1d, 2d (for images) 3d (model) or con (console). Image list is best at 1D to start with. Control points listed below images with two items to create a point and create a distance.  There are a few ways to make GCPs... 

  • drag from the list onto the image. (Click and hold on the point in the image to zoom in).
  • click on a point in one image and then click on the other image to place a matching point
  • hold down CTRL key and click and drag point from a 3D view onto an image
  • with control point tool active, clicking in an empty space on an image creates a new point.

Control point tool can be toggled with F3

Link to comment
Share on other sites

Ozbigben - Are you in any way affiliated with Reality Capture?  It's fine for you to participate here, but we ask that you disclose your relationship to the company.

 

I ask because you have a history of tweets cheerleading about this software and repeatedly saying it's a "game changer"

 

As to your specific post, it isn't a very useful comparison of the software without more data.  As Tom points out there are many options for alignment in Photoscan, and we don't know how RC is doing the alignment.  If you choose a lower level of accuracy when you do alignment in Photoscan, and also whether you use pair preselection can drastically affect the alignment time.  Also the time alone isn't the only driving factor for comparing software.  At CHI, we are driven by scientific approaches and we want to look at and be able to optimize data by reducing the errors.  There are many metrics for this including reprojection uncertainty.

 

I am not suggesting that one product is better than another, only that without more information and a clear understanding of the software options chosen and the metrics, the assessment can't be done.

 

Carla

Link to comment
Share on other sites

Ozbigben - Are you in any way affiliated with Reality Capture?  It's fine for you to participate here, but we ask that you disclose your relationship to the company.

 

Hi Carla

 

Fair question since this was also my first post here.  I am not affiliated with them in any way. I'm a Technical Support Officer at the University of Melbourne http://digitisation.unimelb.edu.au/  ...and I'm also active on the Photoscan forum (bigben).  I'm a scientific photographer by qualification (BAppSci Photography, RMIT) and have been exploring photogrammetry for a couple of years now to provide support and training for staff at our university.  You can see some of my experiments here: https://sketchfab.com/uomdigitisation/models  and I also participate in the Cultural Heritage lounge of Sketchfab's forum. 

 

It was a pretty big call to make so early on in an evaluation, but I am now continuing experiments that I had previously abandoned because I had reach the limits of our hardware.

 

[edit]I also agree with your comments on metrics and I don't know enough on that yet to be able to provide meaningful comment on RC.  Given the performance though, I think it is worth in depth investigation.

 

Cheers

 

Ben

Link to comment
Share on other sites

Thanks for clarifying your background and (non) relationship to Reality Capture.  I agree it is worth looking at, and that's why folks are doing so and discussing it here.  Personally, I don't want to put a lot of time into something where the actual pricing and licensing structure haven't been announced.

 

Carla

Link to comment
Share on other sites

  • 2 months later...

I'm currently running the basic licence if anyone wants any info/ comparisons.  I'm sure many will baulk at the idea of a subscription licence, especially when compared to Photoscan's academic price... Crossing my fingers that some other option may be available for academic/cultural heritage licencing in the near future.   A lot of my experimentation over the last year with photogrammetry has involved the use of fisheye lenses since the release of support in Photoscan. RC doesn't have a true fisheye distortion model (yet), but I have been able to coax it to produce some good results with the distortion models it does have.  e.g. Large scene including several detailed sculptures: http://files.digitisation.unimelb.edu.au/potree/pointclouds/rc-fisheye-test2.html I'll be mentoring a student doing a project on this area over the year which will include a comparison of Photoscan and Reality Capture, including an estimated timeline for producing 3D models of all sculptures.

Link to comment
Share on other sites

Archived

This topic is now archived and is closed to further replies.

×
×
  • Create New...