def generate_mesh(self):
"""
Generate a mesh from self.data and saves it.
:return:
"""
self.make_network_input()
mesh = self.network.module.generate_mesh(self.data.network_input)
path = '/'.join([self.opt.training_media_path, str(self.flags.media_count)]) + ".obj"
image_path = '/'.join([self.data.image_path, '00.png'])
mesh_processor.save(mesh, path, self.colormap)
self.flags.media_count += 1
return {"output_path": path,
"image_path": image_path}
def demo(self, demo_path, input_path_points=None):
"""
This function takes an image or pointcloud path as input and save the mesh infered by Atlasnet
Extension supported are ply npy obg and png
:return: path to the generated mesh
"""
ext = demo_path.split('.')[-1]
self.data = self.datasets.dataset_train.load(demo_path)
self.data = EasyDict(self.data)
if input_path_points is None:
input_path_points = demo_path
#prepare normalization
get_normalization = self.datasets.dataset_train.load(input_path_points)
get_normalization = EasyDict(get_normalization)
self.make_network_input()
mesh = self.network.module.generate_mesh(self.data.network_input)
if get_normalization.operation is not None:
# Undo any normalization that was used to preprocess the input.
vertices = torch.from_numpy(mesh.vertices).clone().unsqueeze(0)
get_normalization.operation.invert()
unnormalized_vertices = get_normalization.operation.apply(vertices)
mesh = pymesh.form_mesh(
vertices=unnormalized_vertices.squeeze().numpy(),
faces=mesh.faces)
if self.opt.demo:
path = demo_path.split('.')
path[-2] += "AtlasnetReconstruction"
path[-1] = "ply"
path = ".".join(path)
else:
path = '/'.join(
[self.opt.training_media_path,
str(self.flags.media_count)]) + ".ply"
self.flags.media_count += 1
print(f"Atlasnet generated mesh at {path}!")
mesh_processor.save(mesh, path, self.colormap)
return path
