Open source point cloud classification
-
- I have made 40-50 posts
- Posts: 44
- Joined: Fri Mar 23, 2018 4:23 pm
- 6
- Full Name: AJS
- Company Details: Nomad
- Company Position Title: Surveyor
- Country: UK
- Linkedin Profile: No
Open source point cloud classification
does anyone know of any os projects similar to 'https://pointly.ai/'?
-
- I have made 40-50 posts
- Posts: 44
- Joined: Fri Mar 23, 2018 4:23 pm
- 6
- Full Name: AJS
- Company Details: Nomad
- Company Position Title: Surveyor
- Country: UK
- Linkedin Profile: No
Re: Open source point cloud classification
http://www.cloudcompare.org/doc/wiki/in ... O_(plugin)
I found this neat plugin. i'm going to give this a go for now.
I found this neat plugin. i'm going to give this a go for now.
- smacl
- Global Moderator
- Posts: 1409
- Joined: Tue Jan 25, 2011 5:12 pm
- 13
- Full Name: Shane MacLaughlin
- Company Details: Atlas Computers Ltd
- Company Position Title: Managing Director
- Country: Ireland
- Linkedin Profile: Yes
- Location: Ireland
- Has thanked: 627 times
- Been thanked: 657 times
- Contact:
Re: Open source point cloud classification
I'd recommend having a look at Florent Poux's work on segmentation, see http://pointcloudproject.com/the-future ... rspective/ It has a big advantage of not requiring training data so potentially a very efficient workflow and precursor to other automated feature extraction techniques. Not sure if he has any publicly available tools in the open source domain as yet but I'd guess they're coming. The issue with training a neural net against terrestrial scanned point cloud data is that there are a lot of inconsistencies in how a given object appears based on distance from the scanner and point density. I'd be interested in hearing how you get on with Canupo.sudo_ki wrote: ↑Sun Aug 23, 2020 11:08 am does anyone know of any os projects similar to 'https://pointly.ai/'?
-
- I have made 40-50 posts
- Posts: 44
- Joined: Fri Mar 23, 2018 4:23 pm
- 6
- Full Name: AJS
- Company Details: Nomad
- Company Position Title: Surveyor
- Country: UK
- Linkedin Profile: No
Re: Open source point cloud classification
nice rea.
I'm tempted to read his thesis. https://orbi.uliege.be/handle/2268/235520
I can' see anything on github for now, but let's wait and see.
There are so many interesting things happening right now, I can't wait to see more algorithms in the reality capture space.
Will let you know how i get on. My main goal is to find something i can use to batch process and remove vegetation.
I'm tempted to read his thesis. https://orbi.uliege.be/handle/2268/235520
I can' see anything on github for now, but let's wait and see.
There are so many interesting things happening right now, I can't wait to see more algorithms in the reality capture space.
Will let you know how i get on. My main goal is to find something i can use to batch process and remove vegetation.
- smacl
- Global Moderator
- Posts: 1409
- Joined: Tue Jan 25, 2011 5:12 pm
- 13
- Full Name: Shane MacLaughlin
- Company Details: Atlas Computers Ltd
- Company Position Title: Managing Director
- Country: Ireland
- Linkedin Profile: Yes
- Location: Ireland
- Has thanked: 627 times
- Been thanked: 657 times
- Contact:
Re: Open source point cloud classification
You're welcome to a demo license to SCC which is pretty good at bare earth extraction and subsequent object heighting, but is neither free nor command line. I'd also have a look at LAStools which can be batch driven and so far as I know is free for non-commercial use.sudo_ki wrote: ↑Tue Aug 25, 2020 1:30 pm nice rea.
I'm tempted to read his thesis. https://orbi.uliege.be/handle/2268/235520
I can' see anything on github for now, but let's wait and see.
There are so many interesting things happening right now, I can't wait to see more algorithms in the reality capture space.
Will let you know how i get on. My main goal is to find something i can use to batch process and remove vegetation.
It is certainly an interesting area, you could spend years just working through all through the research material with new stuff arriving all the time.
-
- I have made 40-50 posts
- Posts: 44
- Joined: Fri Mar 23, 2018 4:23 pm
- 6
- Full Name: AJS
- Company Details: Nomad
- Company Position Title: Surveyor
- Country: UK
- Linkedin Profile: No
Re: Open source point cloud classification
Thank you. https://www.youtube.com/watch?v=miyEQj36Ew0 this looks really good!You're welcome to a demo license to SCC which is pretty good at bare earth extraction and subsequent object heighting, but is neither free nor command line.
XD sometimes i feel like i do more reading than workcould spend years just working through all through the research material
-
- V.I.P Member
- Posts: 958
- Joined: Sun Nov 01, 2009 11:18 pm
- 14
- Full Name: Dennis Hirota
- Company Details: Sam O Hirota Inc
- Company Position Title: President
- Country: USA
- Linkedin Profile: Yes
- Location: Hawaii, USA
- Has thanked: 87 times
- Been thanked: 379 times
Re: Open source point cloud classification
Shanesmacl wrote: ↑Mon Aug 24, 2020 10:18 amI'd recommend having a look at Florent Poux's work on segmentation, see http://pointcloudproject.com/the-future ... rspective/ It has a big advantage of not requiring training data so potentially a very efficient workflow and precursor to other automated feature extraction techniques. Not sure if he has any publicly available tools in the open source domain as yet but I'd guess they're coming. The issue with training a neural net against terrestrial scanned point cloud data is that there are a lot of inconsistencies in how a given object appears based on distance from the scanner and point density. I'd be interested in hearing how you get on with Canupo.sudo_ki wrote: ↑Sun Aug 23, 2020 11:08 am does anyone know of any os projects similar to 'https://pointly.ai/'?
I do not have much time, but here are some examples of the CC plugin Canupo running on my Ryzen 3990X. I can tell you that this example of running 128 threads does makes a difference in time vs. Dual Xeon at 16 threads. Tweeking the resampling also makes a significant difference in time, the larger minimum distance the faster compute time since there are fewer core points to deal with. This looks like significant manual labor to get the configuration files in a decent shape. It would be handy if there are meter based files available.
The Otira files are available from the CC forum if you look hard.
You do not have the required permissions to view the files attached to this post.
- Poux
- I have made 30-40 posts
- Posts: 31
- Joined: Wed Nov 05, 2014 9:24 pm
- 9
- Full Name: Florent POUX
- Company Details: 3D Academy - UTwente
- Company Position Title: Course Director
- Country: France
- Linkedin Profile: Yes
- Location: France
- Been thanked: 4 times
- Contact:
Re: Open source point cloud classification
Thanks for the share Shane!smacl wrote: ↑Mon Aug 24, 2020 10:18 amI'd recommend having a look at Florent Poux's work on segmentation, see http://pointcloudproject.com/the-future ... rspective/ It has a big advantage of not requiring training data so potentially a very efficient workflow and precursor to other automated feature extraction techniques. Not sure if he has any publicly available tools in the open source domain as yet but I'd guess they're coming. The issue with training a neural net against terrestrial scanned point cloud data is that there are a lot of inconsistencies in how a given object appears based on distance from the scanner and point density. I'd be interested in hearing how you get on with Canupo.sudo_ki wrote: ↑Sun Aug 23, 2020 11:08 am does anyone know of any os projects similar to 'https://pointly.ai/'?
Indeed, they are coming soon . In the meantime, all the theoretical background is open access (https://scholar.google.com/citations?us ... AAAJ&hl=en), thus replicating should not be hard for those with the skills to do so.
What is currently super robust is the unsupervised segmentation part (getting nice clusters), and the classification really depends on the model at hand (if you have a lot of training data, I recommend SPG or RF - if you have little to no training data I recommend going through self-learning, see:
https://www.linkedin.com/posts/florent- ... 77888-X4Fc)
The expectation should be that it recognizes rightly 90% of elements in your scene, and find 90% of them. Tested for indoor point clouds with 13 classes, and currently in test for a wide range of applications. For those interested, you can PM me via linkedin and I can put your data (.las file) to the test.
1. Example segmentation with noisy point cloud 2. Example segmentation with mass-market sensor (Matterport) 3. Classification results Cheers and Stay safe,
Adj. Prof. Dr.-Ing Florent Poux
Point Cloud Lab
Jack Dangermond Award Press Release
You do not have the required permissions to view the files attached to this post.
Florent POUX
https://learngeodata.eu
https://learngeodata.eu
-
- I have made 40-50 posts
- Posts: 44
- Joined: Fri Mar 23, 2018 4:23 pm
- 6
- Full Name: AJS
- Company Details: Nomad
- Company Position Title: Surveyor
- Country: UK
- Linkedin Profile: No