LiDAR training data

[bogdan-ardelean]

Hello!

I've downloaded LAZ files from https://zenodo.org/record/3459803#.X4BcWy-w29a. In some of them, individual trees aren't labeled.
For example:

NEON_D17_TEAK_DP1_315000_4094000_classified_point_cloud_colorized_crop.laz
NEON_D08_LENO_DP1_383000_3523000_classified_point_cloud_colorized.laz
NEON_D08_LENO_DP1_383000_3523000_classified_point_cloud_colorized_crop.laz
NEON_D03_OSBS_DP1_405000_3286000_classified_point_cloud.laz
NEON_D03_OSBS_DP1_405000_3286000_classified_point_cloud_colorized.laz

SJER, NIWO and MLBS files contain label data as described in this repo.

Am I missing something?

Thanks!

[bw4sz]

That's right. The benchmark is still in flux. The paper is about to be submitted. We dropped the 3D labels since we didn't think many people would use them. Those three sites were the original sites. What's your use case? Is still worthwhile given the very sparse LiDAR density? Most applications require alot more points than NEON gets with their sensor. The evaluation code won't cover reading in laz files and labeling them but i'm happy to just run the routine to label the existing images for the 'image-annotated crowns'. See https://travis-ci.org/github/weecology/NeonTreeEvaluation_package for a bit more recent detail. If you give me a sense for how they can be used I can check with the co-authors. Provided it doesn't take me a ton of time I can make it all open source. If not we can talk about data sharing/authorship. I don't think that'll be needed, I should have all the scripts around, but just letting you know. I"m out of the office, but I can check it out next week. Best, Ben

…

On Fri, Oct 9, 2020 at 6:29 AM bogdan-ardelean ***@***.***> wrote: Hello! I've downloaded LAZ files from https://zenodo.org/record/3459803#.X4BcWy-w29a. In some of them, individual trees aren't labeled. For example: NEON_D17_TEAK_DP1_315000_4094000_classified_point_cloud_colorized_crop.laz NEON_D08_LENO_DP1_383000_3523000_classified_point_cloud_colorized.laz NEON_D08_LENO_DP1_383000_3523000_classified_point_cloud_colorized_crop.laz NEON_D03_OSBS_DP1_405000_3286000_classified_point_cloud.laz NEON_D03_OSBS_DP1_405000_3286000_classified_point_cloud_colorized.laz SJER, NIWO and MLBS files contain label data as described in this repo. Am I missing something? Thanks! — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#30>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJHBLDI4O22RTJBQ5UYWZTSJ4F5HANCNFSM4SKCUVHQ> .

-- Ben Weinstein, Ph.D. Postdoctoral Fellow University of Florida http://benweinstein.weebly.com/

[bogdan-ardelean]

Thanks for the reply. I thought that somehow they were left out by mistake.
I was just curios about applying some point cloud classification nets on them.
I read your paper on DeepForest and you also mentioned PointNet/PointNet++. Have you ever got a change to try them on this data? You're right, since the data is sparse I would also suggest that it may not be worthwhile going into that much trouble.

We're doing some research for an application to address illegal logging detection and prevention. A part of illegal logging happens on parcels during legal logging.
This is major problem in Romania, and in some other Eastern European countries, so we want to provide a tool for Romsilva, a state-owned enterprise responsible for dealing with the protection of the forests, so that they can timely monitor those parcels and also have an up to date inventory. We're exploring solutions based on UAVs mounted with LiDAR or RGB cameras and satellite imagery with spatial resolution of 30cm. It would be nice if we could identify individual trees, so we're exploring hybrid approaches in order to increase accuracy and reduce costs.

Also, we would like to share the data that will result from this project in order to stimulate the community in finding new solutions and use cases.

If you're interested in our project, we can discuss about it. If you're ok with it, I can leave you a message on email.

Have a nice day!
Bogdan

[bw4sz]

*Always happy to chat. I'm going to close this issue. If anyone wants to labeled the point cloud, I can get around to it, but My sense is that the really sparse data will need some serious work to be usable with pointnets. A problem for the far future.*

On Sat, Oct 10, 2020 at 4:22 AM bogdan-ardelean ***@***.***> wrote: Thanks for the reply. I thought that somehow they were left out by mistake. I was just curios about applying some point cloud classification nets on them. I read your paper on DeepForest and you also mentioned PointNet/PointNet++. Have you ever got a change to try them on this data? You're right, since the data is sparse I would also suggest that it may not be worthwhile going into that much trouble.

We're doing some research for an application to address illegal logging

…

detection and prevention. A part of illegal logging happens on parcels during legal logging. This is major problem in Romania, and in some other Eastern European countries, so we want to provide a tool for Romsilva, a state-owned enterprise responsible for dealing with the protection of the forests, so that they can timely monitor those parcels and also have an up to date inventory. We're exploring solutions based on UAVs mounted with LiDAR or RGB cameras and satellite imagery with spatial resolution of 30cm. It would be nice if we could identify individual trees, so we're exploring hybrid approaches in order to increase accuracy and reduce costs. Also, we would like to share the data that will result from this project in order to stimulate the community in finding new solutions and use cases. If you're interested in our project, we can discuss about it. If you're ok with it, I can leave you a message on email. Have a nice day! Bogdan — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#30 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJHBLB2JT3OM42AU7TVT3DSKA7YRANCNFSM4SKCUVHQ> .

-- Ben Weinstein, Ph.D. Postdoctoral Fellow University of Florida http://benweinstein.weebly.com/