joaquinsalas
commented
2 years ago Dear Ben, nice to hear from you.
Thanks for making us aware of your development of DeepForest. Thanks for making your code available!
We also reported a tree crowd tree detector in @article{pulido2020assessment, title={Assessment of tree detection methods in multispectral aerial images}, author={Pulido, Dagoberto and Salas, Joaqu{\'\i}n and R{\"o}s, Matthias and Puettmann, Klaus and Karaman, Sertac}, journal={Remote Sensing}, volume={12}, number={15}, pages={2379}, year={2020}, publisher={Multidisciplinary Digital Publishing Institute} }
but we missed making public the code.
Within DeepForest, I was particularly impressed by your work counting birds. It seems that your results may be in pair with current SOTA methods. Congratulations.
We have not tried the dataset that you pointed out. It will be exciting and we will be happy to share the results with you.
Currently, we are assessing a deep learning-based tree detector and have not tried parameterizing the algorithm from one forest and applying it to another. But we will be happy to keep talking with you about whether there are effective ways to run this experiment.
Best regards.
Joaquin
On Sat, Dec 4, 2021 at 4:13 PM Ben Weinstein @.***> wrote:
Hello,
I have alot of experience with ALS data for forestry (i'm the developer here: https://deepforest.readthedocs.io/), but i've never been able to successfully get good representations from sparse LiDAR point data. NEON ( https://www.neonscience.org/data-collection/lidar) collects LiDAR data for large forest plots across the US. I've tried all the standard approaches in the R LiDR package without alot of success. I sense that this is mostly due to the sparse point density of the ALS cloud. An average of 5 points per meter is decent, but it can be weaker in some spots. Have far you tested this tool in these conditions? Have you tried parameterizing the algorithm from one forest and applying it to another?
Best,
Ben Weinstein
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/porteratzo/TreeTool/issues/2, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACLU4SCII6N2LSDKBESA3GDUPKHBZANCNFSM5JMA2JEQ . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.
--
Joaquin Salas
bw4sz
Hello,
I have alot of experience with ALS data for forestry (i'm the developer here: https://deepforest.readthedocs.io/), but i've never been able to successfully get good representations from sparse LiDAR point data. NEON (https://www.neonscience.org/data-collection/lidar) collects LiDAR data for large forest plots across the US. I've tried all the standard approaches in the R LiDR package without alot of success. I sense that this is mostly due to the sparse point density of the ALS cloud. An average of 5 points per meter is decent, but it can be weaker in some spots. Have far you tested this tool in these conditions? Have you tried parameterizing the algorithm from one forest and applying it to another?
Best,
Ben Weinstein