As you use (in this implementation) a fixed voxel size, if you have a sparse zone, you may get no triangles. In that case, you need to adapt the algorithm to have its voxel size adapt to the local context.
Dear Dr. Poux, I hope you're doing well! First of all, thanks a lot for your amazing and useful materials-they really motivate me to keep learning. I’m very interested in 3D reconstruction, and it’s not just a hobby-it’s closely related to my career. Your videos have helped me a lot, and I appreciate them so much. But I’m facing a challenge and would love your advice! I’m trying to create a 3D mesh of eyeglasses, but because they’re transparent and shiny, it feels almost impossible to model them. My current approach is to turn the shiny frames into a binary image, fill in the white areas (light reflections) with a photo editor, and then use 3D reconstruction methods. Unfortunately, the results aren’t great. Do you have any suggestions for modeling objects like this? I would really appreciate your help! Thanks again for all the great content. Best regards,
thanks a lot for your kind words! beautiful project you have indeed! my workflow recommendation , to follow your idea, which is super nice, would be to indeed create masks on the areas that you have to reconstruct with photogrammetry (glasses) how accurate do you need it? Based on this, you can then generate glasses with genAI in place using constraints form the relevant points of the fixtures
Hi Florent. Really interesting video. How is this process different from creating a 3d mesh from a pointcloud in an application like RealityCapture if any? Ive been using registered single .e57 files to create a mesh. I gues im trying to understand if that software is using a similar process or if what you are describing is a different process. Thanks!
thanks! great question! so usually, meshing in photogrammetry software is guided by images, thus slightly different than this approach, and more linked with delaunay.
Thanks for such a well explained video👍
Glad it was helpful!
It's very interesting and informative video❤
Glad you liked it
Where can we find the files you are using in the tutorial?
on the academy website (tutorial seciton) learngeodata.eu
Does this algorithm require good point cloud quality? what if it is a bit sparse and at some place incomplete?
As you use (in this implementation) a fixed voxel size, if you have a sparse zone, you may get no triangles. In that case, you need to adapt the algorithm to have its voxel size adapt to the local context.
Dear Dr. Poux,
I hope you're doing well! First of all, thanks a lot for your amazing and useful materials-they really motivate me to keep learning.
I’m very interested in 3D reconstruction, and it’s not just a hobby-it’s closely related to my career. Your videos have helped me a lot, and I appreciate them so much. But I’m facing a challenge and would love your advice!
I’m trying to create a 3D mesh of eyeglasses, but because they’re transparent and shiny, it feels almost impossible to model them. My current approach is to turn the shiny frames into a binary image, fill in the white areas (light reflections) with a photo editor, and then use 3D reconstruction methods. Unfortunately, the results aren’t great.
Do you have any suggestions for modeling objects like this? I would really appreciate your help!
Thanks again for all the great content.
Best regards,
thanks a lot for your kind words! beautiful project you have indeed! my workflow recommendation , to follow your idea, which is super nice, would be to indeed create masks on the areas that you have to reconstruct with photogrammetry (glasses)
how accurate do you need it? Based on this, you can then generate glasses with genAI in place using constraints form the relevant points of the fixtures
Hello sir.
How to compare different point cloud architecture for their performance parameters.please
great question! I put that to the todo list :).
Great video! Can I accomplish the same thing with Apple’s Metal Programming instead of python?
yep this is feasible!
Hi Florent. Really interesting video. How is this process different from creating a 3d mesh from a pointcloud in an application like RealityCapture if any? Ive been using registered single .e57 files to create a mesh. I gues im trying to understand if that software is using a similar process or if what you are describing is a different process. Thanks!
thanks! great question! so usually, meshing in photogrammetry software is guided by images, thus slightly different than this approach, and more linked with delaunay.
why its far from reality so how pro 3D scanners get the right mesh i already see the point clouds pretty so what is wrong ?
Hey! I am not sure I understand the question :). Could you rephrase maybe?
ineed the code
Code is in the course 3D Segmentor OS if you want to use it commercially, or on the Medium article