File size: 9,920 Bytes
6df8db0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""Phase 5: Material & Texture Module.

Generates:
- PBR materials (albedo, metallic, roughness, normal)
- Texture baking from multi-view images
- Lighting estimation for relightable scenes
"""

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 MaterialTextureModule(nn.Module):
    """Generate PBR materials and bake textures onto meshes."""
    
    def __init__(
        self,
        model_size: str = "L",
        device: str = "cuda",
        dtype: torch.dtype = torch.float16,
        use_pbr: bool = True,
        cache_dir: Optional[str] = None,
    ):
        super().__init__()
        self.model_size = model_size
        self.device = device
        self.dtype = dtype
        self.use_pbr = use_pbr
        self.cache_dir = cache_dir
        
        # Material generation model (placeholder for now)
        self._material_model = None
        
        # Material type priors
        self.material_priors = {
            "wall": {"albedo": [0.9, 0.9, 0.9], "metallic": 0.0, "roughness": 0.8},
            "floor_wood": {"albedo": [0.6, 0.4, 0.2], "metallic": 0.0, "roughness": 0.6},
            "floor_tile": {"albedo": [0.8, 0.8, 0.8], "metallic": 0.1, "roughness": 0.3},
            "floor_carpet": {"albedo": [0.5, 0.3, 0.2], "metallic": 0.0, "roughness": 0.9},
            "ceiling": {"albedo": [0.95, 0.95, 0.95], "metallic": 0.0, "roughness": 0.9},
            "furniture_wood": {"albedo": [0.5, 0.3, 0.15], "metallic": 0.0, "roughness": 0.5},
            "furniture_fabric": {"albedo": [0.6, 0.5, 0.4], "metallic": 0.0, "roughness": 0.8},
            "furniture_leather": {"albedo": [0.4, 0.2, 0.1], "metallic": 0.1, "roughness": 0.4},
            "furniture_metal": {"albedo": [0.7, 0.7, 0.7], "metallic": 0.9, "roughness": 0.2},
            "furniture_plastic": {"albedo": [0.8, 0.8, 0.8], "metallic": 0.0, "roughness": 0.3},
            "furniture_glass": {"albedo": [0.9, 0.9, 0.9], "metallic": 0.0, "roughness": 0.05},
            "default": {"albedo": [0.7, 0.7, 0.7], "metallic": 0.0, "roughness": 0.5},
        }
    
    def generate_room_materials(
        self,
        room_shell_mesh: "trimesh.Trimesh",  # type: ignore
        image: Image.Image,
        semantic_seg: np.ndarray,
    ) -> "trimesh.Trimesh":  # type: ignore
        """
        Generate materials for room shell (walls, floor, ceiling).
        
        Uses semantic segmentation to determine material types
        and input image for color extraction.
        """
        if not self.use_pbr:
            return room_shell_mesh
        
        # Extract dominant colors from image regions
        img_np = np.array(image)
        
        # Determine material types from semantic segmentation
        floor_region = semantic_seg == 1
        ceiling_region = semantic_seg == 2
        wall_regions = (semantic_seg == 3) | (semantic_seg == 4)
        
        # Extract colors from corresponding image regions
        floor_color = self._extract_dominant_color(img_np, floor_region)
        ceiling_color = self._extract_dominant_color(img_np, ceiling_region)
        wall_color = self._extract_dominant_color(img_np, wall_regions)
        
        # Create materials
        floor_mat = self._create_material("floor_wood", color=floor_color)
        ceiling_mat = self._create_material("ceiling", color=ceiling_color)
        wall_mat = self._create_material("wall", color=wall_color)
        
        # Apply materials to mesh faces
        # In practice, this would be done per-face based on which room part the face belongs to
        # For now, store materials as mesh metadata
        room_shell_mesh.materials = {
            "floor": floor_mat,
            "ceiling": ceiling_mat,
            "walls": wall_mat,
        }
        
        return room_shell_mesh
    
    def generate_object_materials(
        self,
        object_mesh: "trimesh.Trimesh",  # type: ignore
        multiviews: List[Image.Image],
        object_info: Dict,
    ) -> Tuple["trimesh.Trimesh", List[Dict]]:  # type: ignore
        """
        Generate PBR materials for a furniture object.
        
        Uses multi-view images to bake texture and infer material properties.
        """
        if not self.use_pbr:
            return object_mesh, []
        
        class_name = object_info.get("class_name", "furniture")
        
        # Infer material type from class and image analysis
        material_type = self._infer_material_type(class_name, multiviews[0])
        
        # Extract dominant color from multi-view images
        colors = [self._extract_dominant_color(np.array(mv), np.ones((mv.size[1], mv.size[0]), dtype=bool))
                  for mv in multiviews]
        avg_color = np.mean(colors, axis=0)
        
        # Create material
        material = self._create_material(material_type, color=avg_color)
        
        # Create simple UV atlas texture
        texture = self._bake_texture(object_mesh, multiviews, material)
        
        # Attach texture to mesh
        if texture is not None:
            object_mesh.visual = object_mesh.visual.to_texture()
            # In production, set actual texture image
            object_mesh.material_override = material
        
        materials = [material]
        
        return object_mesh, materials
    
    def estimate_lighting(
        self,
        image: Image.Image,
    ) -> Dict:
        """
        Estimate scene lighting from input image.
        
        Returns:
            {
                "environment_map": HDR environment map (placeholder),
                "key_light_direction": [x, y, z],
                "key_light_intensity": float,
                "fill_light_intensity": float,
                "ambient_intensity": float,
                "color_temperature": float,  # Kelvin
            }
        """
        img_np = np.array(image)
        
        # Simple heuristic lighting estimation
        # In production, use trained lighting estimation network
        
        # Estimate brightness
        brightness = img_np.mean()
        
        # Estimate color temperature from average color
        avg_color = img_np.mean(axis=(0, 1))
        
        # Warm = more red, Cool = more blue
        color_temp = 6500  # Default daylight
        if avg_color[2] > avg_color[0] * 1.2:
            color_temp = 8000  # Cool
        elif avg_color[0] > avg_color[2] * 1.2:
            color_temp = 3000  # Warm
        
        # Estimate light direction from shadows
        # Placeholder: assume light from top-left
        light_dir = np.array([0.3, 0.8, 0.2])
        light_dir = light_dir / np.linalg.norm(light_dir)
        
        return {
            "environment_map": None,  # Would generate HDR probe
            "key_light_direction": light_dir.tolist(),
            "key_light_intensity": float(brightness / 255.0 * 2.0),
            "fill_light_intensity": float(brightness / 255.0 * 0.5),
            "ambient_intensity": float(brightness / 255.0 * 0.3),
            "color_temperature": float(color_temp),
        }
    
    def _extract_dominant_color(
        self,
        image: np.ndarray,
        mask: np.ndarray,
    ) -> np.ndarray:
        """Extract dominant color from image region."""
        if mask.sum() == 0:
            return np.array([0.7, 0.7, 0.7])
        
        masked_pixels = image[mask]
        
        # K-means-ish: use median for robustness
        dominant_color = np.median(masked_pixels, axis=0) / 255.0
        
        return dominant_color
    
    def _create_material(
        self,
        material_type: str,
        color: Optional[np.ndarray] = None,
    ) -> Dict:
        """Create PBR material from type and color."""
        prior = self.material_priors.get(material_type, self.material_priors["default"])
        
        if color is not None:
            albedo = color.tolist()
        else:
            albedo = prior["albedo"]
        
        return {
            "type": material_type,
            "albedo": albedo,
            "metallic": prior["metallic"],
            "roughness": prior["roughness"],
            "normal_scale": 1.0,
            "ao_scale": 1.0,
            # Texture maps (would be actual textures in production)
            "albedo_map": None,
            "metallic_map": None,
            "roughness_map": None,
            "normal_map": None,
            "ao_map": None,
        }
    
    def _infer_material_type(
        self,
        class_name: str,
        image: Image.Image,
    ) -> str:
        """Infer material type from object class and visual appearance."""
        class_lower = class_name.lower()
        
        # Map class to material type
        material_map = {
            "sofa": "furniture_fabric",
            "chair": "furniture_fabric",
            "table": "furniture_wood",
            "coffee_table": "furniture_wood",
            "bed": "furniture_fabric",
            "desk": "furniture_wood",
            "bookshelf": "furniture_wood",
            "lamp": "furniture_metal",
            "wardrobe": "furniture_wood",
            "tv_stand": "furniture_wood",
            "rug": "floor_carpet",
        }
        
        return material_map.get(class_lower, "furniture_wood")
    
    def _bake_texture(
        self,
        mesh: "trimesh.Trimesh",  # type: ignore
        multiviews: List[Image.Image],
        material: Dict,
    ) -> Optional[Image.Image]:
        """
        Bake multi-view images into a unified UV texture.
        
        Uses visibility-aware projection to handle occlusions.
        """
        # Placeholder: in production, this would be proper UV unwrapping + projection
        # For now, return the first multi-view as the texture
        
        if multiviews:
            return multiviews[0]
        return None