Agisoft Metashape is pretty good as well. Results are comparable to RealityCapture (depending a bit on what you want) but personal license is only 179$.
I tried all sorts of free photogrammetry softwares way back but after I tried Metashape I decided to buy a license. With 32 GB of memory it's great. I'm taking photogrammetry models as a hobby - here's a recent one made with Metashape https://skfb.ly/6S6NR
Btw, super shameless plug: If you're looking for 3D capture where photogrammetry doesn't work well (indoors, complex environments) you may find Dot3D (https://www.dotproduct3d.com/dot3dpro.html) quite helpful. It uses depth sensors and works on any recent Android or Windows device. You can get a 14 day free trial and we're happy to extend if needed.
Yes, we're getting there. Of course a > $50k laser scanner will still outperform it in terms of accuracy but the new generation of depth sensors (e.g. Intel RealSense L515) is really quite good. We're going to announce support for this very soon and will also put some datasets online for comparison.
Hey, that sounds potentially relevant to my employer. I'm a small fry in our org but would like to get some of our heavy hitters to take a closer look at what you guys have. Who should we contact (if folks find this relevant)? I prefer not to link my private online persona to my professional role but the page https://www.dotproduct3d.com/workflows.html has several references to our products already so you likely are aware of my employer :)
I would echo the recommendation, but add that you really also need a good GPU. All the steps are highly parallelisable and take significantly longer on a CPU only.
Agisoft has a relatively simple interface, but there's a lot of complexity underneath if you need to tweak things.
This is great! Do you have any resources (youtube videos, articles, etc) for how to do this with a drone? I got outside trying to take pictures of my house but I wasn't sure how many photos to take to get a good render.
Same rules of thumb apply to both drone and handheld cameras. With a drone you can capture large wider areas, though, which is really cool (a city block... a small mountain... whatever).
The 3d reproduction quality is roughly to what you provide in photos-meaning if you view your model from a similar camera distance relatively as the source image it will look good - or further away.
For a high resolution scan you likely want to have a hierachical set of pictures, First a roundabout of say 32 pics.
Then similar turn around of the details that you want to model clearly like columns, stairs, and whatnot protruding and receding from the house facade.
Prefer a bright overcast day - it's way better than a direct sunlight.
Overall rule of thumb is that when you move, consequent picks should have 80% overlap
I usually notice only after taking the photos that I missed an angle and some details I wanted to see are not there. No prob. Since you took the images on overcast day just go back and take the extra few picks missing.
For full on neurotic mode, Put camera on manual, take RAW images, and keep your aperture, shutter speed and ISO constant throughout the shoot.
Even though you have a drone, you still likely want to take photos of ground details by hand (unless you have a real good camera on the drone - my 12 Mp really is noticeably worse than the 24 Mp - in this case you want all the pixels as you can as that helps with the alignment - just as long as they are NOT mushed by poor JPG compression - like my Mavic does-hence RAW:s).
There are lots of tips in archaeology blogs, I think, as well.
For reference, check out the stuff around open source packages of visualsfm, colmap and meshroom, as well as the relevant online reaources of the software you are using.
Don't want to speak for the OP, but that's typically the kind of quality I get directly out of Meshroom with no work other than throwing the pics in it and waiting ~ 4 hours (less if you have a fast, modern GPU)
"with no work other than throwing the pics in it and waiting ~ 4 hours (less if you have a fast, modern GPU)"
Personally I find taking the photos (~1000) is more work than actually waiting the software to complete the processing. Sure, the processing takes longer, but it's not as if you are turning the crank on the PC the whole time :)
Biggest job was sourcing the images. That took maybe two hours. Of course PC had to compute some hours, but I can do other things while that is going on so not really a time investment as such (do it over the night).
Roughly 800. Half by drone (Mavic Pro) with 12 MP camera and half on foot with Sony Alpha 6000 (24 MP) with fixed lens. Had to take low shots on foot mainly because 24 MP picks are much better source data than 12 MP picks. That's why the roof details are not as good as the details on ground.
I haven't tried the commercial packages, but Meshroom is really impressive.
With enough drone pics of a building, it's pretty much:
- launch meshroom
- throw pics in there
- press the "run" button
- come back 3/4 hours later
- load hi-rez fully textured model into blender
- bit of cleanup
- render
In particular, in the last step, you can do orthographic renderings that look like architect plans.
I highly recommend Meshroom. In particular, unlike many photogrammetry software I used before, my experience has been that there is almost no tweaking required.
The only minor gripe: the models come out a bit "heavy" by default (in some places, too much geometry).
