Implementing the mesh fitting with PyTorch3D

The input point cloud is contained in pedestrian.ply. The mesh can be visualized using the vis_input.py code snippet. The main code snippet for fitting a mesh model to the point cloud is contained in deform1.py:

1. We will start by importing the needed packages:

   import os
   import sys
   import torch
   from pytorch3d.io import load_ply, save_ply
   from pytorch3d.io import load_obj, save_obj
   from pytorch3d.structures import Meshes
   from pytorch3d.utils import ico_sphere
   from pytorch3d.ops import sample_points_from_meshes
   from pytorch3d.loss import (
       chamfer_distance,
       mesh_edge_loss,
       mesh_laplacian_smoothing,
       mesh_normal_consistency,
   )
   import numpy as np

2. We then declare a PyTorch device. If you have GPUs, then the device would be created to use GPUs. Otherwise, the device has to use CPUs:

   if torch.cuda.is_available():
       device...