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
| """
NVIDIA Cosmos Transfer: 场景变体生成
从单个重建场景生成百万级变体:
- 交通行为变化
- 天气/光照变化
- 传感器配置变化
- 座舱乘员变化
"""
import numpy as np
from dataclasses import dataclass, field
from typing import List, Dict
import torch
@dataclass
class CosmosConfig:
"""Cosmos Transfer 配置"""
base_model: str = "Cosmos-3"
mode: str = "transfer"
variation_axes: Dict = field(default_factory=lambda: {
'weather': ['clear', 'rain', 'fog', 'snow', 'night'],
'lighting': ['dawn', 'noon', 'dusk', 'night', 'tunnel'],
'behavior': ['normal', 'aggressive', 'drowsy', 'distracted'],
'occupant': ['adult', 'child', 'infant', 'empty', 'pet'],
'traffic': ['sparse', 'normal', 'dense', 'congested'],
})
n_variations: int = 1000
physics_validation: bool = True
output_format: str = "OpenUSD"
def generate_cabin_variations(base_scene: dict,
config: CosmosConfig) -> List[dict]:
"""
从基础座舱场景生成变体
Args:
base_scene: 基础场景参数
config: Cosmos 配置
Returns:
variations: 变体场景列表
"""
variations = []
for i in range(config.n_variations):
var = {
'scene_id': f"var_{i:04d}",
'weather': np.random.choice(config.variation_axes['weather']),
'lighting': np.random.choice(config.variation_axes['lighting']),
'behavior': np.random.choice(config.variation_axes['behavior']),
'occupant': np.random.choice(config.variation_axes['occupant']),
'traffic': np.random.choice(config.variation_axes['traffic']),
'physics': {
'gravity': 9.81,
'friction': np.random.uniform(0.3, 0.8),
'temperature': np.random.uniform(-10, 40),
}
}
variations.append(var)
return variations
class CosmosReasonValidator:
"""
Cosmos Reason: 物理合理性验证
过滤掉物理不可能的生成场景
"""
def __init__(self):
self.validation_rules = {
'gravity_check': True,
'collision_check': True,
'object_permanence': True,
'causality': True,
}
def validate(self, generated_scene: dict) -> bool:
"""
验证生成场景的物理合理性
Returns:
valid: 是否通过物理验证
"""
if not self.validation_rules['gravity_check']:
return False
return True
class CabinDataGenerator:
"""
座舱合成数据生成器
目标: 为 DMS/OMS/CPD/OOP 生成训练数据
"""
SCENARIOS = {
'DMS': {
'fatigue': ['yawning', 'head_nod', 'eye_closure', 'microsleep'],
'distraction': ['phone', 'eating', 'talking', 'adjusting'],
'emotion': ['angry', 'sad', 'happy', 'neutral', 'stressed'],
'impairment': ['alcohol_mild', 'alcohol_severe', 'medication'],
},
'OMS': {
'occupant': ['adult_driver', 'adult_passenger', 'child_seat',
'infant_seat', 'empty', 'pet'],
'posture': ['normal', 'leaning', 'sleeping', 'reaching',
'turned', 'slumped'],
'seatbelt': ['correct', 'incorrect', 'unfastened'],
},
'CPD': {
'infant_seat': ['rear_facing', 'forward_facing'],
'child_age': ['newborn', 'infant', 'toddler', 'child'],
'position': ['left_rear', 'right_rear', 'center_rear'],
'condition': ['sleeping', 'crying', 'moving', 'still'],
},
'OOP': {
'posture': ['forward_lean', 'side_lean', 'backward_lean',
'curled', 'kneeling', 'standing'],
'severity': ['mild', 'moderate', 'severe'],
}
}
def __init__(self, config: CosmosConfig):
self.config = config
self.validator = CosmosReasonValidator()
def generate_batch(self, category: str, n: int = 100) -> List[dict]:
"""
生成指定类别的合成数据
Args:
category: 'DMS', 'OMS', 'CPD', 'OOP'
n: 生成数量
Returns:
samples: 合成数据样本
"""
scenarios = self.SCENARIOS.get(category, {})
samples = []
for i in range(n):
sample = {
'id': f"{category}_{i:04d}",
'category': category,
}
for key, values in scenarios.items():
sample[key] = np.random.choice(values)
sample['environment'] = {
'lighting': np.random.choice(
self.config.variation_axes['lighting']),
'weather': np.random.choice(
self.config.variation_axes['weather']),
}
sample['physics_valid'] = self.validator.validate(sample)
samples.append(sample)
return samples
if __name__ == "__main__":
config = CosmosConfig()
generator = CabinDataGenerator(config)
for category in ['DMS', 'OMS', 'CPD', 'OOP']:
samples = generator.generate_batch(category, n=500)
valid = sum(1 for s in samples if s['physics_valid'])
print(f"{category}: 生成 500, 物理 {valid} ({valid/5:.0f}%)")
s = samples[0]
print(f" 示例: {s.get('fatigue', s.get('posture', s.get('child_age', '')))}")
base_scene = {'location': 'highway', 'time': 'noon'}
variations = generate_cabin_variations(base_scene, config)
print(f"\n基础场景变体: {len(variations)} 个")
print(f"天气分布: {dict(zip(*np.unique([v['weather'] for v in variations], return_counts=True)))}")
|