Material to get started - deep learning algorithm based point cloud classification of eCognition 10.2
Hi, the next week I will start using the deep learning algorithm for the classification of a point cloud in an urban area. Does anyone of you have tutorials, papers, manuals, procedures to follow, etc. to get started? Because I have not found any material neither from Trimble nor from third parties on this deep learning algorithm of eCognition 10.2.
There are only 2 presentation videos on Youtube (https://www.youtube.com/watch?v=m7uyeHjrQRU and https://www.youtube.com/watch?v=SmybG-tt-nc) and the "Reference Book" (paragraph 9.9) about the deep learning algorith for point cloud classification, but they are not useful.
Do you know how it works the deep learning algorith for point cloud classification? Is it based on a pre-trained model?
Thank you
Was this article helpful?
3 comments
Hi,
there is free eCognition training material on our Learn.Trimble.com platform. Here is a direct link to our deep learning free training material --> https://learn.trimble.com/learn/course/external/view/elearning/3698/from-the-ground-up-deep-learning
For Point Cloud classification we have a newly implemented deep learning classification algorithm called "deep learning based point cloud classification (prototype)". This is a prototype but here you can choose from two different Deep Learning models "terrestrial" or "aerial". These are pre-trained models that you can directly apply on your point cloud. Currently you can not create samples or train a model on a Point Cloud directly.
However you can use CNNs and create samples, models etc. within eCognition easily (please see link to the free training). Create samples --> Create model --> train model --> apply model.
Hope that helps?
Cheers,
Matthias
Hi,
thank you for your reply and for the shared material.
Do you know if there is the possibility to ask Trimble to have a mini-course on how the deep learning algorithm works for the classification of point clouds? Or whether manuals or guides will be published on the subject? Or if the functionalities of this tool will be implemented in the future, perhaps not based on a pre-trained model?
I had already seen the material you shared with me but was more interested in direct classification of point clouds.
Thank you
Alberto for the Geomatics lab of the Politecnico di Torino
Hi Alberto,
We are looking into this, currently it is not possible to create your own DL model to classify point clouds. Please post this as a feature request in the Community Wishlist and this will get pushed.
Cheers,
Matthias