search

ErlerPhilipp / points2surf

Learning Implicit Surfaces from Point Clouds (ECCV 2020)
https://www.cg.tuwien.ac.at/research/publications/2020/erler-2020-p2s/
MIT License
463 stars 48 forks source link

Cannot get correct surface representation by using Marchin_Cube #19

Closed QtEngineer closed 2 years ago

QtEngineer commented 2 years ago

Hi, I tried to save query_pts and query_dist of my own network after training,and then used fuction sdf.implicit_surface_to_mesh_directory from point2surf to reconstruct shapes but found that the generated mesh has a lot of vertics and faces (like this pic) I checked query_dist file and found the dists are highly smaller than ABC_train datasets. The average query_dist of mine is 0.00165,while ABC_train dataset is -0.08. I don't know whether too small query_dist will cause reconstructing failed. How can I fix it？ 20220518013127

ErlerPhilipp commented 2 years ago

Are you using the provided code for reconstruction? For example eval_p2s_vanilla.sh which calls full_eval.py. You can adapt these bash script to use your own model. If you use it correctly, you should get all the outputs in the results directory.

Even if the absolute distances are completely wrong, the main source of error is the occupancy. A large number of vertices and faces is normal for marching cubes. There is no mesh decimation applied in my code.

QtEngineer commented 2 years ago

Thanks a lot ！