File size: 14,772 Bytes
dcb20f6 | 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 | """Phase 4: Scene Assembly Module.
Optimizes room layout, resolves collisions, normalizes scale,
and builds the editable scene graph representation.
"""
from typing import Dict, List, Optional, Tuple
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
class SceneAssemblyModule(nn.Module):
"""Assemble individual objects into a coherent room scene."""
def __init__(
self,
device: str = "cuda",
dtype: torch.dtype = torch.float16,
):
super().__init__()
self.device = device
self.dtype = dtype
# Furniture dimension priors (meters) for scale normalization
self.furniture_priors = {
"sofa": {"width": 2.0, "depth": 0.9, "height": 0.8},
"chair": {"width": 0.5, "depth": 0.5, "height": 0.9},
"table": {"width": 1.2, "depth": 0.8, "height": 0.75},
"coffee_table": {"width": 1.0, "depth": 0.6, "height": 0.45},
"bed": {"width": 2.0, "depth": 1.5, "height": 0.5},
"desk": {"width": 1.4, "depth": 0.7, "height": 0.75},
"bookshelf": {"width": 1.0, "depth": 0.3, "height": 2.0},
"lamp": {"width": 0.3, "depth": 0.3, "height": 1.5},
"wardrobe": {"width": 1.5, "depth": 0.6, "height": 2.1},
"tv_stand": {"width": 1.2, "depth": 0.4, "height": 0.5},
"rug": {"width": 2.0, "depth": 1.5, "height": 0.02},
"plant": {"width": 0.3, "depth": 0.3, "height": 1.0},
"furniture": {"width": 0.8, "depth": 0.8, "height": 0.8}, # default
}
def assemble(
self,
room_shell_mesh: "trimesh.Trimesh", # type: ignore
object_meshes: List["trimesh.Trimesh"], # type: ignore
room_layout: Dict,
detected_objects: Dict,
depth_map: np.ndarray,
) -> Dict:
"""
Assemble room scene from individual components.
Steps:
1. Place objects at detected positions
2. Normalize scales using furniture priors
3. Ensure objects rest on floor
4. Resolve collisions
5. Build scene graph
6. Merge into unified mesh
"""
# Step 1: Initial placement from detected positions
placed_objects = self._place_objects(
object_meshes, detected_objects, room_layout
)
# Step 2: Scale normalization
normalized_objects = self._normalize_scales(
placed_objects, detected_objects, depth_map
)
# Step 3: Gravity constraint (objects on floor)
grounded_objects = self._apply_gravity(
normalized_objects, room_layout
)
# Step 4: Collision detection and resolution
resolved_objects = self._resolve_collisions(
grounded_objects, room_layout
)
# Step 5: Build scene graph
scene_graph = self._build_scene_graph(
resolved_objects, room_layout, detected_objects
)
# Step 6: Merge into unified mesh
scene_mesh = self._merge_scene(
room_shell_mesh, resolved_objects
)
return {
"scene_mesh": scene_mesh,
"object_meshes": resolved_objects,
"scene_graph": scene_graph,
}
def _place_objects(
self,
object_meshes: List["trimesh.Trimesh"], # type: ignore
detected_objects: Dict,
room_layout: Dict,
) -> List["trimesh.Trimesh"]: # type: ignore
"""Place objects at their detected positions in 3D space."""
placed = []
floor_height = room_layout.get("floor", {}).get("height", 0.0)
for i, mesh in enumerate(object_meshes):
if i in detected_objects:
obj_info = detected_objects[i]
bbox = obj_info.get("bbox", [0, 0, 100, 100])
depth_range = obj_info.get("depth_range", [1.0, 3.0])
# Compute 3D position from bbox center + depth
# Simple approximation: center of bbox at mean depth
img_h, img_w = depth_map.shape if 'depth_map' in locals() else (512, 512)
x1, y1, x2, y2 = bbox
cx = (x1 + x2) / 2
cy = (y1 + y2) / 2
mean_depth = np.mean(depth_range)
# Convert image coordinates to 3D
# Assume camera at origin, looking down -z
fx = fy = max(img_w, img_h)
cx_cam = img_w / 2
cy_cam = img_h / 2
x_3d = (cx - cx_cam) * mean_depth / fx
z_3d = mean_depth # depth is z in camera frame
# Position mesh
mesh_copy = mesh.copy()
# Center mesh
centroid = mesh_copy.centroid if hasattr(mesh_copy, 'centroid') else mesh_copy.bounds.mean(axis=0)
mesh_copy.apply_translation([-centroid[0], 0, -centroid[2]])
# Move to detected position
mesh_copy.apply_translation([x_3d, floor_height, z_3d])
placed.append(mesh_copy)
else:
placed.append(mesh.copy())
return placed
def _normalize_scales(
self,
object_meshes: List["trimesh.Trimesh"], # type: ignore
detected_objects: Dict,
depth_map: np.ndarray,
) -> List["trimesh.Trimesh"]: # type: ignore
"""Normalize object scales using furniture priors and depth."""
normalized = []
for i, mesh in enumerate(object_meshes):
mesh_copy = mesh.copy()
# Get class name
class_name = "furniture"
if i in detected_objects:
class_name = detected_objects[i].get("class_name", "furniture")
# Get prior dimensions
prior = self.furniture_priors.get(
class_name, self.furniture_priors["furniture"]
)
# Compute current dimensions
bounds = mesh_copy.bounds
current_dims = bounds[1] - bounds[0]
# Compute scale factors
# Use largest dimension for scale reference
max_current = max(current_dims)
max_prior = max(prior["width"], prior["depth"], prior["height"])
if max_current > 0.001: # Avoid division by zero
scale_factor = max_prior / max_current
# Apply non-uniform scaling to match prior
target_scale = np.array([
prior["width"] / max(current_dims[0], 0.001),
prior["height"] / max(current_dims[1], 0.001),
prior["depth"] / max(current_dims[2], 0.001),
])
# Clamp scale to reasonable range
scale_factor = np.clip(scale_factor, 0.1, 3.0)
target_scale = np.clip(target_scale, 0.1, 3.0)
# Use uniform scale for stability
mesh_copy.apply_scale(scale_factor)
normalized.append(mesh_copy)
return normalized
def _apply_gravity(
self,
object_meshes: List["trimesh.Trimesh"], # type: ignore
room_layout: Dict,
) -> List["trimesh.Trimesh"]: # type: ignore
"""Ensure all objects rest on the floor."""
floor_height = room_layout.get("floor", {}).get("height", 0.0)
grounded = []
for mesh in object_meshes:
mesh_copy = mesh.copy()
# Find lowest point
if len(mesh_copy.vertices) > 0:
min_y = mesh_copy.vertices[:, 1].min()
# Move so lowest point is at floor height
delta_y = floor_height - min_y
mesh_copy.apply_translation([0, delta_y, 0])
grounded.append(mesh_copy)
return grounded
def _resolve_collisions(
self,
object_meshes: List["trimesh.Trimesh"], # type: ignore
room_layout: Dict,
) -> List["trimesh.Trimesh"]: # type: ignore
"""Detect and resolve inter-object collisions."""
resolved = list(object_meshes)
max_iterations = 50
for iteration in range(max_iterations):
collisions_found = False
for i in range(len(resolved)):
for j in range(i + 1, len(resolved)):
try:
# Check collision
collision = resolved[i].collision_manager
is_collision = False # Placeholder
# Simple bounding box collision test
b1 = resolved[i].bounds
b2 = resolved[j].bounds
overlap = (
b1[0][0] < b2[1][0] and b1[1][0] > b2[0][0] and
b1[0][1] < b2[1][1] and b1[1][1] > b2[0][1] and
b1[0][2] < b2[1][2] and b1[1][2] > b2[0][2]
)
if overlap:
collisions_found = True
# Push apart along smallest overlap axis
overlaps = [
min(b1[1][0] - b2[0][0], b2[1][0] - b1[0][0]),
min(b1[1][1] - b2[0][1], b2[1][1] - b1[0][1]),
min(b1[1][2] - b2[0][2], b2[1][2] - b1[0][2]),
]
min_axis = np.argmin(overlaps)
push_dir = np.zeros(3)
push_dir[min_axis] = 1.0
# Push in opposite directions
push_dist = overlaps[min_axis] * 0.5 + 0.05
center_i = resolved[i].bounds.mean(axis=0)
center_j = resolved[j].bounds.mean(axis=0)
if center_i[min_axis] < center_j[min_axis]:
resolved[i].apply_translation(-push_dir * push_dist)
resolved[j].apply_translation(push_dir * push_dist)
else:
resolved[i].apply_translation(push_dir * push_dist)
resolved[j].apply_translation(-push_dir * push_dist)
except Exception:
pass
if not collisions_found:
break
return resolved
def _build_scene_graph(
self,
object_meshes: List["trimesh.Trimesh"], # type: ignore
room_layout: Dict,
detected_objects: Dict,
) -> Dict:
"""Build editable scene graph from assembled objects."""
nodes = []
edges = []
# Room shell node
nodes.append({
"id": "room_shell",
"type": "room",
"label": "room",
"bbox": None,
})
# Object nodes
for i, mesh in enumerate(object_meshes):
class_name = "furniture"
if i in detected_objects:
class_name = detected_objects[i].get("class_name", "furniture")
center = mesh.bounds.mean(axis=0)
dims = mesh.bounds[1] - mesh.bounds[0]
nodes.append({
"id": i,
"type": "object",
"label": class_name,
"position": center.tolist(),
"dimensions": dims.tolist(),
"mesh_index": i,
})
# Edge: object is IN room
edges.append({
"from": i,
"to": "room_shell",
"relation": "in",
})
# Infer spatial relationships between objects
for i in range(len(object_meshes)):
for j in range(i + 1, len(object_meshes)):
center_i = object_meshes[i].bounds.mean(axis=0)
center_j = object_meshes[j].bounds.mean(axis=0)
dist = np.linalg.norm(center_i - center_j)
# Proximity threshold
if dist < 2.0:
# Determine relationship
if abs(center_i[1] - center_j[1]) < 0.1:
relation = "next_to"
elif center_i[1] > center_j[1] + 0.2:
relation = "on"
else:
relation = "near"
edges.append({
"from": i,
"to": j,
"relation": relation,
"distance": float(dist),
})
return {
"nodes": nodes,
"edges": edges,
}
def _merge_scene(
self,
room_shell_mesh: "trimesh.Trimesh", # type: ignore
object_meshes: List["trimesh.Trimesh"], # type: ignore
) -> "trimesh.Trimesh": # type: ignore
"""Merge room shell and objects into unified scene mesh."""
import trimesh
meshes = [room_shell_mesh] + list(object_meshes)
# Filter out empty meshes
valid_meshes = [m for m in meshes if hasattr(m, 'vertices') and len(m.vertices) > 0]
if not valid_meshes:
return trimesh.Trimesh()
try:
scene_mesh = trimesh.util.concatenate(valid_meshes)
except Exception:
# Fallback: add meshes one by one
scene_mesh = valid_meshes[0]
for m in valid_meshes[1:]:
try:
scene_mesh += m
except Exception:
pass
return scene_mesh
def reassemble_with_textures(
self,
room_shell_mesh: "trimesh.Trimesh", # type: ignore
textured_objects: List["trimesh.Trimesh"], # type: ignore
scene_graph: Dict,
) -> "trimesh.Trimesh": # type: ignore
"""Re-assemble scene with textured objects."""
return self._merge_scene(room_shell_mesh, textured_objects)
|