search

yifita / DSS

Differentiable Surface Splatting
441 stars 31 forks source link

Visualization of the results #24

Closed pbonazzi closed 2 years ago

pbonazzi commented 2 years ago

Hi , I have a couple of questions on the results I am getting during training .

1) The directory val contains best.ply which if visualized with open3d with the following code, is showing an empty view. 

import open3d as o3d
path="best.ply"
textured_mesh = o3d.io.read_point_cloud(path)
o3d.visualization.draw_geometries([textured_mesh])

2) I can see how the model trains using tensorboard . And I have noticed there are points surrounding the point cloud , is this expected ?

3) The directory vis is empty. Why ?

4) In addition, how can I get the final mesh (the grey one you show in the paper) ?

5) Finally, does the model handle texture in some way ?

lxxue commented 2 years ago

  1. The optimized points are saved in best.ply. I can visualize the point cloud in meshlab. Probably you need to check how open3d visualizes a point cloud.

  2. This is quite common to me. Some points will be pushed to the empty space and stay there.

  3. The visualization is put in tensorboard now. You can check trainer.visualize function for more details and tweak yourself if you want to visualize other stuff.

  4. I am not sure how the mesh was generated in the paper. I think you can use the screened Poisson reconstruction method to get a mesh from the point cloud and the normals.

  5. I think you can render textured images by changing some codes here: https://github.com/yifita/DSS/blob/97d4bfe73ad054197f604c4e3f759e0f90745696/scripts/create_mvr_data_from_mesh.py#L128-L129

pbonazzi commented 2 years ago

Amazing support thanks !