Point Cloud to Mesh based on CUDA
Budget: €18 – €36 EUR
Point Cloud to Mesh based on CUDA:
The project needs to be written in C++.
We need a parallel surface reconstruction algorithm that runs entirely on GPU.
Requirements:
Package should not be licensed under any commercial license. it should be free for DigiSwiss to use and change and distribute without any limitation in functionality or time limit.
Package should be able to reconstruct mesh for 3d point cloud in size of few thousand up to ~1.5 M for variable number of point cloud with different size, in proper time. As example for 8 scanned object with two point cloud around 1M and 6 point cloud with 400K, 500K point per each in less than 8 second.
As input package will receive 3 vector of Eigen::Vecto3f for colour, normal and positions. This is fixed and any internal conversion must be optimized. Result must be also in same format of vector of Eigen::Vector3i as faces. As example open3d mesh can be used as referenced.
Out push mesh must be manifold mesh.
Isotropic remeshing would be an advantage (second priority).
Accuracy of generated mesh must be comparable with current mesh algorithm.
Algorithm must have proper API to define prefer number of vertices, edge, smoothness, and etc to control parameters of the mesh.
Output number if vertex and faces are important for us to reduce the output file size properly down with keeping the accuracy high.
Algorithm should be able to cover filling hole with different size, small or big with uniform faces.
The project needs to be written in C++.
We need a parallel surface reconstruction algorithm that runs entirely on GPU.
Requirements:
Package should not be licensed under any commercial license. it should be free for DigiSwiss to use and change and distribute without any limitation in functionality or time limit.
Package should be able to reconstruct mesh for 3d point cloud in size of few thousand up to ~1.5 M for variable number of point cloud with different size, in proper time. As example for 8 scanned object with two point cloud around 1M and 6 point cloud with 400K, 500K point per each in less than 8 second.
As input package will receive 3 vector of Eigen::Vecto3f for colour, normal and positions. This is fixed and any internal conversion must be optimized. Result must be also in same format of vector of Eigen::Vector3i as faces. As example open3d mesh can be used as referenced.
Out push mesh must be manifold mesh.
Isotropic remeshing would be an advantage (second priority).
Accuracy of generated mesh must be comparable with current mesh algorithm.
Algorithm must have proper API to define prefer number of vertices, edge, smoothness, and etc to control parameters of the mesh.
Output number if vertex and faces are important for us to reduce the output file size properly down with keeping the accuracy high.
Algorithm should be able to cover filling hole with different size, small or big with uniform faces.