Spaces:
Running
on
Zero
Running
on
Zero
File size: 6,909 Bytes
f876753 fc44d4b f876753 fc44d4b f876753 fc44d4b f876753 fc44d4b f876753 fc44d4b f876753 fc44d4b f876753 fc44d4b f876753 fc44d4b f876753 fc44d4b f876753 fc44d4b f876753 fc44d4b f876753 fc44d4b f876753 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 |
from dataclasses import dataclass
import torch
import torch.nn as nn
import os
import numpy as np
from .saving import SaverMixin
from ..utils.mesh import Mesh
from ..utils.general_utils import scale_tensor
@dataclass
class ExporterOutput:
save_name: str
save_type: str
params: dict
class IsosurfaceHelper(nn.Module):
points_range = (0, 1)
@property
def grid_vertices(self):
raise NotImplementedError
class DiffMarchingCubeHelper(IsosurfaceHelper):
def __init__(
self,
resolution,
point_range = (0, 1)
):
super().__init__()
self.resolution = resolution
self.points_range = point_range
from diso import DiffMC
self.mc_func = DiffMC(dtype=torch.float32)
self._grid_vertices = None
self.register_buffer(
"_dummy", torch.zeros(0, dtype=torch.float32), persistent=False
)
@property
def grid_vertices(self):
if self._grid_vertices is None:
# keep the vertices on CPU so that we can support very large resolution
x, y, z = (
torch.linspace(*self.points_range, self.resolution),
torch.linspace(*self.points_range, self.resolution),
torch.linspace(*self.points_range, self.resolution),
)
x, y, z = torch.meshgrid(x, y, z, indexing="ij")
verts = torch.stack([x, y, z], dim=-1).reshape(-1, 3)
verts = verts * (self.points_range[1] - self.points_range[0]) + self.points_range[0]
self._grid_vertices = verts
return self._grid_vertices
def forward(
self,
level,
deformation = None,
isovalue=0.0,
):
level = level.view(self.resolution, self.resolution, self.resolution)
if deformation is not None:
deformation = deformation.view(self.resolution, self.resolution, self.resolution, 3)
v_pos, t_pos_idx = self.mc_func(level, deformation, isovalue=isovalue)
v_pos = v_pos * (self.points_range[1] - self.points_range[0]) + self.points_range[0]
# TODO: if the mesh is good
return Mesh(v_pos=v_pos, t_pos_idx=t_pos_idx)
def isosurface(
space_cache,
forward_field,
isosurface_helper,
):
# the isosurface is dependent on the space cache
# randomly detach isosurface method if it is differentiable
# get the batchsize
if torch.is_tensor(space_cache): #space cache
batch_size = space_cache.shape[0]
elif isinstance(space_cache, dict): #hyper net
# Dict[str, List[Float[Tensor, "B ..."]]]
for key in space_cache.keys():
batch_size = space_cache[key][0].shape[0]
break
# scale the points to [-1, 1]
points = scale_tensor(
isosurface_helper.grid_vertices.to(space_cache.device),
isosurface_helper.points_range,
[-1, 1], # hard coded isosurface_bbox
)
# get the sdf values
sdf_batch, deformation_batch = forward_field(
points[None, ...].expand(batch_size, -1, -1),
space_cache
)
# get the isosurface
mesh_list = []
# check if the sdf is empty
# for sdf, deformation in zip(sdf_batch, deformation_batch):
for index in range(sdf_batch.shape[0]):
sdf = sdf_batch[index]
# the deformation may be None
if deformation_batch is None:
deformation = None
else:
deformation = deformation_batch[index]
# special case when all sdf values are positive or negative, thus no isosurface
if torch.all(sdf > 0) or torch.all(sdf < 0):
print(f"All sdf values are positive or negative, no isosurface")
sdf = torch.norm(points, dim=-1) - 1
mesh = isosurface_helper(sdf, deformation)
mesh.v_pos = scale_tensor(
mesh.v_pos,
isosurface_helper.points_range,
[-1, 1], # hard coded isosurface_bbox
)
# TODO: implement outlier removal
# if cfg.isosurface_remove_outliers:
# mesh = mesh.remove_outlier(cfg.isosurface_outlier_n_faces_threshold)
mesh_list.append(mesh)
return mesh_list
def colorize_mesh(
space_cache,
export_fn,
mesh_list,
activation,
):
"""Colorize the mesh using the geometry's export function and space cache.
Args:
space_cache: The space cache containing feature information
export_fn: The export function from geometry that generates features
mesh_list: List of meshes to colorize
Returns:
List[Mesh]: List of colorized meshes
"""
# Process each mesh in the batch
for i, mesh in enumerate(mesh_list):
# Get vertex positions
points = mesh.v_pos[None, ...] # Add batch dimension [1, N, 3]
# Get the corresponding space cache slice for this mesh
if torch.is_tensor(space_cache):
space_cache_slice = space_cache[i:i+1]
elif isinstance(space_cache, dict):
space_cache_slice = {}
for key in space_cache.keys():
space_cache_slice[key] = [
weight[i:i+1] for weight in space_cache[key]
]
# Export features for the vertices
out = export_fn(points, space_cache_slice)
# Update vertex colors if features exist
if "features" in out:
features = out["features"].squeeze(0) # Remove batch dim [N, C]
# Convert features to RGB colors
mesh._v_rgb = activation(features) # Access private attribute directly
return mesh_list
class MeshExporter(SaverMixin):
def __init__(self, save_dir="outputs"):
self.save_dir = save_dir
os.makedirs(save_dir, exist_ok=True)
def get_save_dir(self):
return self.save_dir
def get_save_path(self, filename):
return os.path.join(self.save_dir, filename)
def convert_data(self, x):
if isinstance(x, torch.Tensor):
return x.detach().cpu().numpy()
return x
def export_obj(
mesh,
save_path,
save_normal = False,
):
"""
Export mesh data to OBJ file format.
Args:
mesh_data: Dictionary containing mesh data (vertices, faces, etc.)
save_path: Path to save the OBJ file
Returns:
List of saved file paths
"""
# Create exporter
exporter = MeshExporter(os.path.dirname(save_path))
# Export mesh
save_paths = exporter.save_obj(
os.path.basename(save_path),
mesh,
save_mat=None,
save_normal=save_normal and mesh.v_nrm is not None,
save_uv=False,
save_vertex_color=mesh.v_rgb is not None,
)
return save_paths
|