GGG - Rough or Noisy? Metrics for Noise Detection in SfM Reconstructions
The proposed dataset is part of the paper with the same name "Rough or Noisy? Metrics for Noise Detection in SfM Reconstructions", which proposes 9 geometrical and SfM capturing setup metrics for training a classification method for detecting noise on a SfM reconstructed surfaces. The dataset contains images for Structure from Motion reconstruction of 15 objects. The images were taken using a Canon 5Ds DSLR camera with a resolution of 8688x5792. The images contain EXIF data that can be used to assist the 3D reconstruction. For all objects, except the wind turbine blade the images are 36 in a circle around the object. For the wind turbine blade the images are positioned in 2 vertical bands of 17 images each in a semi-circle. Together with the images there are reconstruction point clouds. These are created using Agisoft Metashape (Photoscan). For easier processing they are separated into: - one file containing vertex positions in X,Y,Z column stacked format, together with color for each vertex position in RGB - one file containing the normals in Nx, Ny, Nz column stacked format - one file containing triangulated faces for the used vertices Finally there are manually annotated ground truth segmentations of the point clouds into noise and not noise for each of the 15 objects.
Steps to reproduce
Follow the Python code steps presented in : https://github.com/IvanNik17/Extract-Features-from-3D-meshes-for-Machine-Learning As well as the Unity project presented in: https://github.com/IvanNik17/Extract-Capturing-Setup-Features