File size: 13,856 Bytes
2033370 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 | """Phase 3: 3D Reconstruction Module.
Reconstructs:
- Room shell (walls, floor, ceiling) as planar meshes
- Per-object 3D meshes using TRELLIS.2 or native InteriorFusion-L
- Scene-level Gaussian Splatting representation
"""
import os
from typing import Dict, List, Optional, Tuple, Union
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
from PIL import Image
class Reconstruction3DModule(nn.Module):
"""Reconstruct 3D geometry from multi-view images."""
def __init__(
self,
model_size: str = "L",
device: str = "cuda",
dtype: torch.dtype = torch.float16,
cache_dir: Optional[str] = None,
):
super().__init__()
self.model_size = model_size
self.device = device
self.dtype = dtype
self.cache_dir = cache_dir
# Lazy load reconstruction models
self._trellis_model = None
self._native_model = None
def reconstruct_room_shell(
self,
room_shell_views: Dict[str, Image.Image],
room_layout: Dict,
depth_map: np.ndarray,
) -> "trimesh.Trimesh": # type: ignore
"""
Reconstruct room shell (walls, floor, ceiling) as planar meshes.
Uses detected layout planes from scene understanding to create
watertight room geometry.
"""
try:
import trimesh
except ImportError:
print("Warning: trimesh not available, using numpy fallback")
return None
meshes = []
# Floor mesh
floor = room_layout.get("floor", {})
if floor:
floor_mesh = self._create_floor_mesh(floor, room_layout)
if floor_mesh is not None:
meshes.append(floor_mesh)
# Ceiling mesh
ceiling = room_layout.get("ceiling", {})
if ceiling:
ceiling_mesh = self._create_ceiling_mesh(ceiling, room_layout)
if ceiling_mesh is not None:
meshes.append(ceiling_mesh)
# Wall meshes
walls = room_layout.get("walls", [])
for wall in walls:
wall_mesh = self._create_wall_mesh(wall, room_layout)
if wall_mesh is not None:
meshes.append(wall_mesh)
# Combine all meshes
if meshes:
try:
room_shell = trimesh.util.concatenate(meshes)
except Exception:
room_shell = meshes[0]
for m in meshes[1:]:
room_shell += m
return room_shell
# Fallback: create simple box room
return self._create_fallback_room(room_layout)
def _create_floor_mesh(self, floor: Dict, room_layout: Dict) -> Optional["trimesh.Trimesh"]: # type: ignore
"""Create floor plane mesh."""
try:
import trimesh
except ImportError:
return None
dims = room_layout.get("dimensions", {})
width = dims.get("width", 5.0)
depth = dims.get("depth", 5.0)
height = floor.get("height", 0.0)
# Create rectangular floor
vertices = np.array([
[-width/2, height, -depth/2],
[width/2, height, -depth/2],
[width/2, height, depth/2],
[-width/2, height, depth/2],
])
faces = np.array([
[0, 1, 2],
[0, 2, 3],
])
mesh = trimesh.Trimesh(vertices=vertices, faces=faces)
# Add UV coordinates for texture mapping
uvs = np.array([
[0, 0],
[1, 0],
[1, 1],
[0, 1],
])
mesh.visual = trimesh.visual.TextureVisuals(uv=uvs)
return mesh
def _create_ceiling_mesh(self, ceiling: Dict, room_layout: Dict) -> Optional["trimesh.Trimesh"]: # type: ignore
"""Create ceiling plane mesh."""
try:
import trimesh
except ImportError:
return None
dims = room_layout.get("dimensions", {})
width = dims.get("width", 5.0)
depth = dims.get("depth", 5.0)
height = ceiling.get("height", 2.7)
vertices = np.array([
[-width/2, height, -depth/2],
[width/2, height, -depth/2],
[width/2, height, depth/2],
[-width/2, height, depth/2],
])
# Ceiling faces point downward
faces = np.array([
[0, 2, 1],
[0, 3, 2],
])
mesh = trimesh.Trimesh(vertices=vertices, faces=faces)
return mesh
def _create_wall_mesh(self, wall: Dict, room_layout: Dict) -> Optional["trimesh.Trimesh"]: # type: ignore
"""Create wall plane mesh."""
try:
import trimesh
except ImportError:
return None
dims = room_layout.get("dimensions", {})
width = dims.get("width", 5.0)
depth = dims.get("depth", 5.0)
height = dims.get("height", 2.7)
normal = np.array(wall.get("normal", [0, 0, 1]))
position = wall.get("position", 0.0)
direction = wall.get("direction", "back")
# Create wall based on direction
if direction in ["back", "front"]:
# Wall perpendicular to z-axis
z = position if direction == "front" else -position
vertices = np.array([
[-width/2, 0, z],
[width/2, 0, z],
[width/2, height, z],
[-width/2, height, z],
])
else: # left or right
# Wall perpendicular to x-axis
x = position if direction == "right" else -position
vertices = np.array([
[x, 0, -depth/2],
[x, 0, depth/2],
[x, height, depth/2],
[x, height, -depth/2],
])
# Determine face orientation based on normal
if normal[2] > 0.5 or normal[0] > 0.5:
faces = np.array([[0, 1, 2], [0, 2, 3]])
else:
faces = np.array([[0, 2, 1], [0, 3, 2]])
mesh = trimesh.Trimesh(vertices=vertices, faces=faces)
return mesh
def _create_fallback_room(self, room_layout: Dict) -> "trimesh.Trimesh": # type: ignore
"""Create a simple box room as fallback."""
import trimesh
dims = room_layout.get("dimensions", {})
width = dims.get("width", 5.0)
depth = dims.get("depth", 5.0)
height = dims.get("height", 2.7)
# Create box with interior
box = trimesh.creation.box(extents=[width, height, depth])
box.apply_translation([0, height/2, 0])
return box
def reconstruct_object(
self,
multiviews: List[Image.Image],
room_layout: Optional[Dict] = None,
depth_map: Optional[np.ndarray] = None,
object_info: Optional[Dict] = None,
) -> Tuple["trimesh.Trimesh", Optional[torch.Tensor]]: # type: ignore
"""
Reconstruct a single furniture object from multi-view images.
Uses TRELLIS.2 for high-quality object reconstruction,
or falls back to simple point cloud reconstruction.
Returns:
(mesh, gaussian_cloud)
"""
# Try TRELLIS.2 if available
mesh = self._try_trellis_reconstruction(multiviews)
if mesh is not None:
return mesh, None
# Fallback: simple reconstruction from depth
return self._fallback_object_reconstruction(multiviews, depth_map, object_info)
def _try_trellis_reconstruction(
self,
multiviews: List[Image.Image],
) -> Optional["trimesh.Trimesh"]: # type: ignore
"""Try to use TRELLIS.2 for object reconstruction."""
try:
# Attempt to import and use TRELLIS
# In production: from trellis import TRELLISPipeline
# For now, placeholder
return None
except ImportError:
return None
def _fallback_object_reconstruction(
self,
multiviews: List[Image.Image],
depth_map: Optional[np.ndarray] = None,
object_info: Optional[Dict] = None,
) -> Tuple["trimesh.Trimesh", Optional[torch.Tensor]]: # type: ignore
"""Simple reconstruction from first multi-view image and depth."""
import trimesh
if depth_map is not None and object_info is not None:
bbox = object_info.get("bbox", [0, 0, 100, 100])
x1, y1, x2, y2 = bbox
# Extract depth region for this object
obj_depth = depth_map[y1:y2, x1:x2]
# Create point cloud from depth
H, W = obj_depth.shape
fx = fy = max(W, H)
cx, cy = W / 2, H / 2
u, v = np.meshgrid(np.arange(W), np.arange(H))
z = obj_depth
x = (u - cx) * z / fx
y = (v - cy) * z / fy
points = np.stack([x, y, z], axis=-1).reshape(-1, 3)
# Remove invalid points
valid = points[:, 2] > 0.1
points = points[valid]
if len(points) > 100:
# Create convex hull as simple mesh
try:
mesh = trimesh.convex.hull_points(points)
return mesh, None
except Exception:
pass
# If hull fails, return point cloud as mesh
if len(points) > 0:
mesh = trimesh.PointCloud(points)
return mesh, None
# Ultimate fallback: small cube
mesh = trimesh.creation.box(extents=[0.5, 0.5, 0.5])
return mesh, None
def build_scene_gaussians(
self,
room_shell_mesh: "trimesh.Trimesh", # type: ignore
object_gaussians: List[Optional[torch.Tensor]],
object_meshes: List["trimesh.Trimesh"], # type: ignore
) -> torch.Tensor:
"""
Build a unified Gaussian Splatting representation for the entire scene.
Converts meshes to Gaussian primitives for fast rendering.
"""
gaussians = []
# Convert room shell mesh to Gaussians
try:
if hasattr(room_shell_mesh, 'vertices') and len(room_shell_mesh.vertices) > 0:
room_gaussians = self._mesh_to_gaussians(room_shell_mesh)
gaussians.append(room_gaussians)
except Exception as e:
print(f"Warning: could not convert room shell to Gaussians: {e}")
# Add per-object Gaussians
for obj_gauss in object_gaussians:
if obj_gauss is not None:
gaussians.append(obj_gauss)
if gaussians:
return torch.cat(gaussians, dim=0)
# Fallback: return empty tensor
return torch.zeros(0, 14, device=self.device)
def _mesh_to_gaussians(
self,
mesh: "trimesh.Trimesh", # type: ignore
num_gaussians_per_face: int = 4,
) -> torch.Tensor:
"""
Convert a mesh to 3D Gaussian primitives.
Each face spawns multiple Gaussians with:
- Position: near face centroid
- Scale: based on face area
- Rotation: aligned with face normal
- Opacity: ~0.9
- Color: from vertex colors or white
"""
if len(mesh.faces) == 0:
return torch.zeros(0, 14, device=self.device)
vertices = torch.tensor(mesh.vertices, dtype=torch.float32, device=self.device)
faces = torch.tensor(mesh.faces, dtype=torch.long, device=self.device)
num_faces = len(faces)
total_gaussians = num_faces * num_gaussians_per_face
# Get face data
v0 = vertices[faces[:, 0]]
v1 = vertices[faces[:, 1]]
v2 = vertices[faces[:, 2]]
# Face centroids
centroids = (v0 + v1 + v2) / 3.0
# Face normals
edges1 = v1 - v0
edges2 = v2 - v0
normals = torch.cross(edges1, edges2, dim=-1)
normals = F.normalize(normals, dim=-1)
# Face areas
areas = 0.5 * torch.norm(normals, dim=-1)
# Build Gaussians
# Gaussian parameters: [x, y, z, scale_x, scale_y, scale_z,
# rot_qx, rot_qy, rot_qz, rot_qw, r, g, b, opacity]
gaussians = []
for i in range(num_gaussians_per_face):
# Offset from centroid
offset = torch.randn_like(centroids) * 0.01
positions = centroids + offset
# Scale based on area
scales = torch.stack([
torch.sqrt(areas) * 0.1 + 0.001,
torch.sqrt(areas) * 0.1 + 0.001,
torch.sqrt(areas) * 0.05 + 0.001,
], dim=-1)
# Rotation from normal
# Simple: identity-ish rotation aligned with normal
rot_identity = torch.tensor([0.0, 0.0, 0.0, 1.0], device=self.device)
rotations = rot_identity.unsqueeze(0).expand(num_faces, -1)
# Color: white default
colors = torch.ones(num_faces, 3, device=self.device) * 0.8
# Opacity
opacity = torch.ones(num_faces, 1, device=self.device) * 0.9
gaussians.append(torch.cat([
positions, scales, rotations, colors, opacity
], dim=-1))
return torch.cat(gaussians, dim=0)
|