3D scanning of as-built with Canvas/Structure Sensor?

[Karl Ottenstein]

Does anyone have direct experience with Occipital's low cost Structure Sensor and Canvas app for iPad?

A friend just pointed me to the Canvas web site:
https://canvas.io
which is a free iPad app (for high power iPads) using the Structure Sensor which is now also owned/made by Occipital. (I dug in and found an earlier version of this was produced by 3D Systems.) Cost is around $400.

This is a 2 camera visual scan that they claim gives accuracy within 1-2%, good enough for rough context for a remodel. Various images/videos on the site.

I downloaded one of the SketchUp models - a scan mesh (they don't create point clouds) can be converted to SU for $29. It cleanly converted to properly grouped ARCHICAD Morphs - windows distinct from walls, e.g. - but the accuracy of elements such as windows, cabinet toe-kicks, etc suggested to me that it was created by manual labor - no doubt somewhere in the world at super low labor rates. I'm a little ethically concerned that Occipital/Canvas doesn't address this likely human factor - or the working conditions for the employees. Anyone know more?

Site shows how accuracy varies based on distance from the iPad because of the way 2-camera depth sensing works. Press release from this fall lists that they bought a LiDAR company to provide more accurate laser measurement solutions as well. Video gives a quick overview of Canvas/Structure Sensor vs Paracosm LiDAR:
https://youtu.be/mogaE8lnUK4

Anyway... any feedback on anyone's experience with Occipital/Canvas appreciated. I'm feeling really tempted for a remodel project coming up...

Thanks,
Karl

[Anonymous]

Hi Karl,

I have a Structure sensor. I made some tests while in beta 19 of Archicad.
It exports obj files I had to translate to e57 through meshlab. e57 is much lighter as format.

I have it attached on my iPad mini 1st edition that the manufacturer claims it will not work. Well it works but when tested on newer tablets with more power I had better capturing results. With stronger machine it captures faster and more accurate.

There is comparison test on the internet that proves that is good enough.

Canvas must be new App haven't tested yet. There was another app before for room scanning.

Occipital also provides you with windows software to attach on laptop computer. Bracket and iPad mini is best solution though it gives you easiness to move around.

Regards

[Karl Ottenstein]

Thanks, ispyridis. It sounds like Occiptal has dramatically changed the software platform with the Canvas app, since it cannot run on older iPads, not even my iPad Air (v1)...and the $29/scan converstion to SketchUp eliminates the need to deal with a mesh.

So, I'm most curious about the current software/hardware usefulness.

In any case, did you find the e57 mesh that you got from your Structure Sensor useful enough for context scans of existing spaces?

Anyone out there who has tried Canvas?

Thanks!
Karl

[Anonymous]

Karl wrote:
In any case, did you find the e57 mesh that you got from your Structure Sensor useful enough for context scans of existing spaces?

Yes e57 is extra light and snap works on it in sections too.

Make sure you use it with latest iPad mini for processor power and mobility.
it exports obj files and mtl. Use mesh lab to convert to e57 or stl although stl too heavy.

Regards

[Anonymous]

Hi all! Alex from Occipital here - I'm the product manager for Canvas. Wanted to address a few questions raised here.

First things first - excited about the interest in Canvas!

Regarding your question about the human aspect, Scan To CAD is a human-assisted service, meaning that it's a mix of both software and people. Our 3D reconstruction (and automated cloud processing) tech does the lion's share of the work, and some of the very CAD-specific work like ensuring the geometry is correctly segmented into components, properly-layered, etc is done by a human to ensure the files you get back are design-ready and professional-grade. I'd love to tell you that AI could do 100% of the whole thing and that files of that level of detail and quality can be produced totally for free in real-time that work perfectly in your architecture or design tool of choice - but not yet! 😉

In any case, it's all done within a very controlled environment, with highly trained professionals, and not farmed out to random CAD outsourcing shops around the world. It's also completely optional, so you can convert the files yourself manually if you prefer.

As ispyridis points out, Canvas does require a newer iPad, as it needs more RAM than the older generations of iPads provide. I do recommend using at least an iPad mini 4, but the newer iPad the better.

Anyway, if you have further questions - please do feel encouraged to add them here and I can respond, or PM me and we can chat further privately. Happy to share some more sample data, talk through use cases, or anything else that would be helpful!

[Karl Ottenstein]

Partial update, in case anyone finds this thread...'

I missed Alex's reply somehow. Excellent to have the product manager appear here so quickly in the conversation!

I bit the bullet and purchased the Structure Sensor and have done some room scans using the scan-to-cad service this week after receiving some excellent support from Alex via Occipital forum, email and phone.

My first scans were returned today and are a mix of high value awesomeness and less than ideal accuracy. Will talk more about it later with some screenshots as I have not had sufficient time do a full accuracy check. (I've seen mostly +/- 0.5% to 1.5% error , but with some 3.5% and even 20% error that I'll talk with Alex about.) This technology cannot be 100% accurate - it's not a $25,000 laser point cloud, so stretching / adjusting is required if a precise as-built is required, which I do want in my case. But it is close enough for well-detailed context if the geometry is adjacent to a tenant-improvement area, specific remodel/renovation area/etc. In the case of context modeling, the imported SketchUp may well be good enough as-is if accurate sections showing building materials aren't needed (etc). Otherwise, ARCHICAD walls would need to be magic-wanded to the imported Morph geometry, and ARCHICAD doors/windows inserted - but the SketchUp-morph will give plan/section/elevation snap points to position those doors/windows quickly for subsequent fine-tuning.

It was pretty awesome that the SketchUp files that were returned to me even modeled cabinetry, some furniture, some complicated trim details and crown moulding.

Equally awesome is the textured mesh that is returned. The scanned mesh is sent to Occipital directly from the iPad app - and upon completion, the scan thumbnails in-app are updated with an icon to let you know the results are ready to download. In-app, you can switch between the original mesh and the cloud-corrected mesh, and also a view that has the full color imagery of the room baked onto the mesh. Measurements can be taken in-app by tapping.

The same photographically textured mesh is available in the computer download zip and can be viewed in the free meshlab app, for a similar virtual reality experience and measurement. The meshlab results are somewhat more cool looking because of screensize, backplane clipping, and the ability to adjust surface lighting. Screenshots to come, eventually.

The computer download zip also contains the SKP file as well as DAE and 2D DWG.

The mesh with baked photographic textures can't be brought into ARCHICAD though - way too many polygons. So, we can't get that kind of 3D context for modeling. But, the SketchUp model works reasonably well. Converting the object to Morphs (one click) then lets you move groups into separate layers/etc.

So far, and considering US labor rates for doing all of this manually, Canvas and the Structure Sensor are looking to me like a reasonable aid in modeling as-builts, or at least renovation / tenant--improvement context. You will need a newer IOS device though. I used a 2017 iPad Pro which has 4 GB of internal memory (the size of the media storage - all that Apple advertises - is not relevant)... and generally consumed more than half of that per room scan (as seen by a progress bar during scanning), and actually ran out of memory on a huge (28' x 32') room, so wouldn't recommend a device with less memory to this community.

Karl

[Anonymous]

Sounds good Karl, would be great to see screenshots when you get a chance. Bit concerned about the accuracy issues though, has Alex commented on this?

[Karl Ottenstein]

Mark wrote:
Sounds good Karl, would be great to see screenshots when you get a chance. Bit concerned about the accuracy issues though, has Alex commented on this?

Alex is patiently waiting for me to send him some screenshots / drawings with measurements to review the errors I saw. Hoping to catch up enough today to get that to him and will post more after he reviews. Thanks!

[Anonymous]

Greetings all, I realize that the last thread on this post ended nearly 2 years ago. But I am looking at the "new and improved" sensor as well. Now that AC 24 has been released is anyone will using this hardware (updated?) and what are you seeing as results? Are you still using the same pipeline to get the objects into ArchiCAD or is there something better you have discovered.
Responses most graciously appreciated.
Cheers,

-Shawn
S-Pod Studio, Los Angeles

[Karl Ottenstein]

I'm glad you posted, Shawn. Canvas' customer 'touch' leaves much to be desired, so I only now found out by going to their site that they now support the new LIDAR (of sorts) enabled iPad Pros, eliminating the need to purchase the Structure Sensor. Since i have an 11" iPad Pro, I just tried it out... and it behaves similarly to an older iPad with the external sensor, perhaps even more smoothly. But, the polygonalization seems cruder. Won't know how this compares until I scan a room and submit it to their 'scan to CAD' service.

But with the free Canvas app, trying this out is now free for anyone with a newer iPad.

The biggest issue with the Structure Sensor itself (vs using the iPad's LIDAR sensor) is the hassle of the mounting brackets, if there is even a bracket available for your particular device.

Accuracy is within the 1-2%, which is often good enough for as-built or context geometry. Hard to beat for the price... but more $$ can equal higher accuracy if you need/want an accurate as-built that doesn't require adjustments to get the dimensions exact. On the other hand, Occipital now advertises that you can submit an as-built drawing/sketch with exact dimensions and they'll make the model match, which would make this the cheapest as-built option out there I think (?) See this page:

https://support.canvas.io/article/198-how-to-submit-manual-dimensions-to-canvas

where the customer surprisingly wants a 47 foot plus wall dimensioned within a 16th of an inch.

Since I last posted and tried Canvas, they now offer Revit/IFC models for $10 more than SketchUp. The SU model was fine for my purposes (reference model). But, I might try a scan at some point with Revit/IFC output to see if those are more convenient for import into AC 24.

I wouldn't purchase the new/improved sensor, but would use the money for a new iPad instead - assuming that there is no other advantage to the new sensor. (The Canvas pages pretty much recommend only getting the sensor if you have old hardware.). Given a new iPad, the Canvas app is free and the only cost would be to evaluate a room scan for $29 for SU or $39 for Revit/IFC:
https://canvas.io/pricing.

Keep in mind that they say that to obtain a reasonably accurate scan/model, you must be standing (as you walk around the space) at most 3 meters / 10 feet from the target surface, making this useful only for those spaces where that is possible. For larger spaces, laser scanners are probably the only option?

If you try it, post what you think.