Prevu3D - Why we took the leap

Discuss all PREVU3D related issues here.
Post Reply
Nicolas Morency
I have made <0 posts
I have made <0 posts
Posts: 5
Joined: Thu Sep 09, 2021 1:51 pm
Full Name: Nicolas Morency
Company Details: Prevu3D
Company Position Title: CEO
Country: Canada
Linkedin Profile: Yes
Has thanked: 4 times
Been thanked: 7 times

Prevu3D - Why we took the leap

Post by Nicolas Morency »

I remember navigating through the forum when I first started looking into 3D scanning.

Back then, capturing colored scans was quite a luxury. There were wild debates as to why someone would even bother capturing imagery when using intensity was plenty for most workflows.

I must admit I was quite thrilled when a new generations of hardware from Faro and Leica came out.

Not only scanning times were much faster, but imagery capture was now accessible.

My goal was to use 3D scanning to minimize modeling efforts, but soon realized dealing with point cloud was a challenge on its own.

I will make a wild statement here, but I think point clouds and imagery are both compromises. I always thought combining the accuracy of points clouds with an imagery overlay was the ultimate output.

Keep in mind my goal was not to reverse engineer CAD files from the scans.

I wanted to skip modeling all together by capturing and generating a textured meshed environment. From there, I could then insert CAD files and start editing my space. My goal was to bring context to design reviews and engage collaboration with non-technical users.

Truth is, apart from us geeks, most people are handed a terrible laptop when they join a company. I was looking for a way to bring the benefits of 3D scanning to those people, the 95% that simply can’t deal with point cloud anyway.

My thought process was simple, how could a cheap XBOX 360 deliver a massive, meshed environment but you need a $3K laptop to deal with a point cloud.

From there, we took the leap to work with meshes and spent the last four years building a solution for them.

We joined the forum to be challenged by the biggest community of laser-scanning experts.

And hopefully, contribute by providing access to a new type of 3D scanning deliverable.
____________________________________________________________________

Video :
youtu.be/8Jv0oBJDgiU

Input: 66GB E57 RTC360
Processing time: 2h33min
Mesh: 120M polygons
Textures : 5620 x 2K
Output size: 4.6GB (Including the app.)
badam
V.I.P Member
V.I.P Member
Posts: 248
Joined: Tue May 11, 2021 5:36 pm
Full Name: Adam Berta
Company Details: InnoScan 3D Hungary Kft
Company Position Title: unknown
Country: Hungary
Linkedin Profile: No
Has thanked: 9 times
Been thanked: 53 times
Contact:

Re: Prevu3D - Why we took the leap

Post by badam »

it is interesting to see this approach but usually the details are lost during meshing. but i'll try it out it would be good to see an on premise solution, because uploads are slow. it would be faster to process localy, and only upload just the result.
User avatar
Leandre Robitaille
V.I.P Member
V.I.P Member
Posts: 288
Joined: Sat Aug 03, 2019 1:53 am
2
Full Name: Leandre Robitaille
Company Details: Cima+
Company Position Title: Civil Technician
Country: Canada
Has thanked: 37 times
Been thanked: 96 times

Re: Prevu3D - Why we took the leap

Post by Leandre Robitaille »

Impressive Nicolas! Happy to see you progress threw out the years. Your mesh conversion looks by far the most accurate scan to mesh conversion out there and results shown are impressive. I don't currently have any use for meshes but threw out the years we have seen a few clients asking for mesh deliverables.Our current solution was to use Cintoo but your results are much better. Will defenetivelly keep your solution in mind for next time.
Is your mesh available in standard formats (fbx/obj/etc) or they can only be viewed threw your app?
User avatar
smacl
Global Moderator
Global Moderator
Posts: 803
Joined: Tue Jan 25, 2011 5:12 pm
10
Full Name: Shane MacLaughlin
Company Details: Atlas Computers Ltd
Company Position Title: Managing Director
Country: Ireland
Linkedin Profile: Yes
Location: Ireland
Has thanked: 329 times
Been thanked: 316 times
Contact:

Re: Prevu3D - Why we took the leap

Post by smacl »

Great looking results Nicholas!
Nicolas Morency
I have made <0 posts
I have made <0 posts
Posts: 5
Joined: Thu Sep 09, 2021 1:51 pm
Full Name: Nicolas Morency
Company Details: Prevu3D
Company Position Title: CEO
Country: Canada
Linkedin Profile: Yes
Has thanked: 4 times
Been thanked: 7 times

Re: Prevu3D - Why we took the leap

Post by Nicolas Morency »

badam wrote: Mon Nov 22, 2021 10:45 pm it is interesting to see this approach but usually the details are lost during meshing. but i'll try it out it would be good to see an on premise solution, because uploads are slow. it would be faster to process localy, and only upload just the result.
Happy to be challenged regarding the resolution, please try it out. Regarding local VS cloud processing, we took the bet to automate the processing online for ease of integration with third parties, access to 16 000 CPU for processing speed and unlimited storage space. For example, when we process a 500GB dataset at full resolution and 12 level of details, the data can easily be multiplied by 5X prior to compression. That kind of disc space is cumbersome for most people. We do have a desktop version we use to develop the pipeline internally, and it might be made available down the road. In the short term, a compression algorithm will be deployed on the SaaS in Q1 2022. We will compress the data up to 85% prior to upload to fasten upload time.
Nicolas Morency
I have made <0 posts
I have made <0 posts
Posts: 5
Joined: Thu Sep 09, 2021 1:51 pm
Full Name: Nicolas Morency
Company Details: Prevu3D
Company Position Title: CEO
Country: Canada
Linkedin Profile: Yes
Has thanked: 4 times
Been thanked: 7 times

Re: Prevu3D - Why we took the leap

Post by Nicolas Morency »

Leandre Robitaille wrote: Tue Nov 23, 2021 3:09 am Impressive Nicolas! Happy to see you progress threw out the years. Your mesh conversion looks by far the most accurate scan to mesh conversion out there and results shown are impressive. I don't currently have any use for meshes but threw out the years we have seen a few clients asking for mesh deliverables.Our current solution was to use Cintoo but your results are much better. Will defenetivelly keep your solution in mind for next time.
Is your mesh available in standard formats (fbx/obj/etc) or they can only be viewed threw your app?
Hey Leandre, it's been a while! Thanks for your comment. You can now export section or even the entire space in various formats : (Vertex Color DXF, XYZ Point Cloud, Textured OBJ, Textured GLB, Textured FBX, STL, PLY). We have on our 2022 roadmap to allow to export the space in standard tiling solutions such as Bentley iTwin and Cesium.
badam
V.I.P Member
V.I.P Member
Posts: 248
Joined: Tue May 11, 2021 5:36 pm
Full Name: Adam Berta
Company Details: InnoScan 3D Hungary Kft
Company Position Title: unknown
Country: Hungary
Linkedin Profile: No
Has thanked: 9 times
Been thanked: 53 times
Contact:

Re: Prevu3D - Why we took the leap

Post by badam »

Nicolas Morency wrote: Tue Nov 23, 2021 4:12 pm
badam wrote: Mon Nov 22, 2021 10:45 pm it is interesting to see this approach but usually the details are lost during meshing. but i'll try it out it would be good to see an on premise solution, because uploads are slow. it would be faster to process localy, and only upload just the result.
Happy to be challenged regarding the resolution, please try it out. Regarding local VS cloud processing, we took the bet to automate the processing online for ease of integration with third parties, access to 16 000 CPU for processing speed and unlimited storage space. For example, when we process a 500GB dataset at full resolution and 12 level of details, the data can easily be multiplied by 5X prior to compression. That kind of disc space is cumbersome for most people. We do have a desktop version we use to develop the pipeline internally, and it might be made available down the road. In the short term, a compression algorithm will be deployed on the SaaS in Q1 2022. We will compress the data up to 85% prior to upload to fasten upload time.
But your 16'000 processing cores will be shared between hundreds of peoples or more. And if you are stuck with an upload speed like 50-100 mbps then it will be a bottleneck. Sometimes you just cannot get better/upload speed but you can easily buy a 32 core amd threadripper with 64 thread. And sometimes you are forced to not upload the data to online services, just local copies are allowed.

If you are working with pointclouds especially if you are working with TLS datas then there is no way you couldn't handle few TB of temporary data. At first you should always keep the original raw data, the registration software native data, and of course the exported e57-s it is just not much more to temporarily keep duplicates. And we are a small company.

The meshes looks great compared to others. (in cloud compare from single setup i created similar result with 1 mm gridded pointcloud and same resolution in poisson reccon, it was not a fast process that is true :) ) Viewer is fast but i for example miss the location of the spheres (i mean you can jump in but don't know where to click to jump in a sphere). And i miss more advanced tools, like section creation, orthoimages. different colorings, like setup cloud, intensity. But it is suprising, that native architectural applications are still faster with pointclouds then meshed object. So until that is changed. it is only good for visualization.

The pricing will be the hardest question of all. Willl it be setup position based or size based, or project based. Will there be any limitations on setup count/project (2000 or more).
Nicolas Morency
I have made <0 posts
I have made <0 posts
Posts: 5
Joined: Thu Sep 09, 2021 1:51 pm
Full Name: Nicolas Morency
Company Details: Prevu3D
Company Position Title: CEO
Country: Canada
Linkedin Profile: Yes
Has thanked: 4 times
Been thanked: 7 times

Re: Prevu3D - Why we took the leap

Post by Nicolas Morency »

badam wrote: Tue Nov 23, 2021 4:41 pm
Nicolas Morency wrote: Tue Nov 23, 2021 4:12 pm
badam wrote: Mon Nov 22, 2021 10:45 pm it is interesting to see this approach but usually the details are lost during meshing. but i'll try it out it would be good to see an on premise solution, because uploads are slow. it would be faster to process localy, and only upload just the result.
Happy to be challenged regarding the resolution, please try it out. Regarding local VS cloud processing, we took the bet to automate the processing online for ease of integration with third parties, access to 16 000 CPU for processing speed and unlimited storage space. For example, when we process a 500GB dataset at full resolution and 12 level of details, the data can easily be multiplied by 5X prior to compression. That kind of disc space is cumbersome for most people. We do have a desktop version we use to develop the pipeline internally, and it might be made available down the road. In the short term, a compression algorithm will be deployed on the SaaS in Q1 2022. We will compress the data up to 85% prior to upload to fasten upload time.
But your 16'000 processing cores will be shared between hundreds of peoples or more. And if you are stuck with an upload speed like 50-100 mbps then it will be a bottleneck. Sometimes you just cannot get better/upload speed but you can easily buy a 32 core amd threadripper with 64 thread. And sometimes you are forced to not upload the data to online services, just local copies are allowed.

If you are working with pointclouds especially if you are working with TLS datas then there is no way you couldn't handle few TB of temporary data. At first you should always keep the original raw data, the registration software native data, and of course the exported e57-s it is just not much more to temporarily keep duplicates. And we are a small company.

The meshes looks great compared to others. (in cloud compare from single setup i created similar result with 1 mm gridded pointcloud and same resolution in poisson reccon, it was not a fast process that is true :) ) Viewer is fast but i for example miss the location of the spheres (i mean you can jump in but don't know where to click to jump in a sphere). And i miss more advanced tools, like section creation, orthoimages. different colorings, like setup cloud, intensity. But it is suprising, that native architectural applications are still faster with pointclouds then meshed object. So until that is changed. it is only good for visualization.

The pricing will be the hardest question of all. Willl it be setup position based or size based, or project based. Will there be any limitations on setup count/project (2000 or more).
Thanks for your note on the mesh.

Since we are on AWS cloud, access to CPU and storage shouldn't be a concern. Agreed for a specific use case such as military, offline processing is a big plus. Regarding the SaaS platform, we are in the process of SOC II compliance to meet the security requirements of most of the large organizations.

You do have more advanced tools on the top menu: View modes such as XRay, Height-mode, Import 2D, 3D CAD files in most industry format, Cut elements in real-time, generate 2D floorplans, 3D mesh export and orthophoto. You can find more info here : https://doc.prevu3d.com/application/basicToolUsage
yankoch
V.I.P Member
V.I.P Member
Posts: 494
Joined: Wed Oct 05, 2011 10:14 am
10
Full Name: Yan Koch
Company Details: TagLabs
Company Position Title: CEO
Country: France
Linkedin Profile: Yes
Has thanked: 35 times
Been thanked: 133 times

Re: Prevu3D - Why we took the leap

Post by yankoch »

One of the big questions is also whether streaming tens of gigabytes of data is desirable given the environmental challenges we face.
There is nothing larger than point clouds or the meshes derived from them.
Streaming a point cloud is Netflix^2.
Nicolas Morency
I have made <0 posts
I have made <0 posts
Posts: 5
Joined: Thu Sep 09, 2021 1:51 pm
Full Name: Nicolas Morency
Company Details: Prevu3D
Company Position Title: CEO
Country: Canada
Linkedin Profile: Yes
Has thanked: 4 times
Been thanked: 7 times

Re: Prevu3D - Why we took the leap

Post by Nicolas Morency »

Here's another demo dataset processed on our server last week :

Partner : Topo3D
Input: 41GB E57
Processing time from Point Cloud to Deliverable: 2h40min
Mesh: 150M polygons and 3600 2K textures
Output size: 4.5GB (Including the app.)


youtu.be/U3K7IWbg2H4
Post Reply

Return to “PREVU3D”