Yes, you can export as .stl file. I am working on another tutorial for parametric modelling

You can use both, but it will not make a huge difference with the current implementations

Great idea! on the roadmap, I will get on that soon 😊

You mean while exporting?

lmao got the same internship assignment???

@@HarryStylesFanSansy Have you any idea about this?

A Very good idea indeed! And this is a vast topic, I will heavily dive on it next month, but basically, decompose the process in (1) classification, and (2) vectorization (regression)

Thanks! Great question! if the downward processes are using the downsampled, then, it should be the pcd_downsampled... But I may have done a dirty trick there, I will get back into the code and reach out thereafter

Most welcome

Hey Millie, thanks a lot for the kind words! Hmmm, it could be a number of reasons, But I think the first step would be to make sure that you can open your ply in another software (E.g. CloudCompare), from there, make sure you have colors, and then, re-export as .ply (binary for example). Then, import into blender and follow the tutorial, maybe adding an 'instance on point" node if things are still not visualized. This should solve your issue

Merci beaucoup!

In a tunnel, I would use a Shape detection approch to isolate the tunnel (E.g RANSAC with Cylinder shape), and then, I would detect the ground with RANSAC again (or CSF), and finally work on isolating the individual remaining elements. What do you think?

Yep, I see I ommitted that, Will do ASAP !

Now in the comments 😊

What capture method did you use? Did you make a movie or took pictures? What camera did you use? What tool did you use for training your data?

Hmm yeah, with some specific shooting and/or material, the result is still not tremendous. Do you want to share your image dataset with me so that I explore a solution?

I support these questions :)

alright i fixed it! on windows I have to insert the complete file path. in DATANAME = "E:/Spyder/Spyder/data/230725_edit.ply" this fixed the reading error.

haha, these are the best bugs, it usually takes some time to get found haha! Thanks for the heads up that you found the solution 😊

Happy about it :)!

CloudCompare :)

Thanks a lot for the kind words! To handle large datasets, if you have memory error, maybe you can try working out of core with memory mapping mechanism. Numpy has some capabilities in this domain, or switching to laspy you have also the ability to load by chunks, which is very helful. And finally, you have data format like parquet that may be a bit lighter on the memory side of things. In production, I tend to use tiling and out of core processing.

Challenging! But I will work on it, on the roadmap of tutorials now ;)

Beautiful idea and process indeed!

Hmmm, indeed, what you can do, is check if your point cloud has color, for example in CloudCompare?

Nerf and Gaussian Splatting are coming up! But here, it stands for Open Software Edition, so it would take it after the creation of the dataset

Thanks a lot! At this stage, you have can add a entity node to get them as particles if this allows you to work on specific single units

Hey Thibault! Sure, Maybe you can explore your dataset in Python or CloudCompare, to check out your color field, and reformat it so that it is rightly taken into account for your Blender approach?

Thanks! Welcome :)! Haha, you have the right mindset, indeed, this is feasible, in one of the course line at the 3D Geodata Academy I show the process on how to do that. But basically, it comes to adding some lines within the loop to extract each segmetn as a single file

Thanks! This is due (from what I read) to your search parameters. For example, if your search condition (radius) takes all the point of your point cloud, this is normal. Try identifying the average distance between points, and set a research condition below this to check out if this solves your issue.

Thanks!

Thanks a lot!

Thanks a lot! Great use case you have! So yes indeed, you can maybe work into a two step proces: data labelling with the help of unsupervised approaches like this workflow; feeding the labelled dataset to Random Forest and PointNet++. You could 1 detect the ground, 2 detect walls, 3 cluster remaining elements, 4 add a conditional rule to give a class to each cluster (if detected from planes, it is a ground if horizontal, else walls; if a cluster has a shape of XXX, it likely is a hosting element of a valve). From that, you check out this video: kzbin.info/www/bejne/qKS6lIdpfbiLnJY , and give each segment as an input to the visualizer to confirm the label. Then the next stage is supervised learning with PointNet++ or other networks (you can check out the 3D Deep Learning Course at learngeodata.eu for that). I hope this helps!

It looks like you have an not correctly formatted ply, try opening it ion CloudCompare, reexport as ply ascci and try again

@@FlorentPoux ok sir.

I have installed anaconda with Spyder IDE but no idea how to import point cloud , where to store them , I have downloaded bin files from symantic kitti. I have seen many of your related videos but not able to sort out the Problem.

let me investigate

Hey! Thanks for your comment! What are you trying to achieve that these tutorials prevent you to do?

Hey! Sure! but I think we have to look at your code to understand better :)

@@FlorentPoux sorry for the late reply . I have got it right later . there was a problem with a function in my code which i rectified it later . thanks for making this content its very useful for the newbies like me .

Great suggestion! In the meantime, You have the PointNet Semantic Segmentation with Code and Model as part of the 3D Deep Learning Course (learngeodata.eu)

Thanks! You have some ways but it is a bit broken (you have to compile from scratch). I would advise getting to work with an isolated environment with Python 3.9

@@FlorentPoux Thanks! It worked. I used Python 3.8

Yes you can! You can use Open3D to do this interactively

You're very welcome!

Hej! The Code is available through the workshop serie, but for the mesh part and the code with another depth estimator, you can get all here: kzbin.info/www/bejne/a6rTgnyLl9V0bcUsi=dj6RvF2Rmx27PDoZ

Yes it would!

@@FlorentPoux ok i have to try , let you know how its goes