# 场景描述:疲劳驾驶员面部特写 prompt = """ A close-up shot of a fatigued driver's face inside a dimly lit vehicle cabin at night. The driver has drooping eyelids with visible redness around the eyes. Yawning with mouth partially open, showing signs of sleep deprivation. The face is illuminated by the faint glow of the dashboard. The driver's head tilts slightly forward, indicating drowsiness. Captured from the driver monitoring camera perspective, front-facing angle. """
result = pipe( prompt=prompt, num_frames=1, height=720, width=1280 )
# 长尾场景:儿童遗留检测模拟 prompt = """ The video begins inside a modern vehicle cabin on a hot summer day. A child in a car seat in the rear seat moves slightly, then becomes still. Sunlight streams through the windows, indicating rising cabin temperature. The scene alternates between exterior shots of the parked vehicle and interior shots of the unresponsive child. No adults visible in the cabin. Time stamps show 10 minutes elapsed, with cabin temperature rising. First-person perspective from the cabin monitoring camera. """
# 机器人操作序列生成 prompt = """ Pick up the red cup from the table and place it on the coaster. This video is captured from a first-person perspective looking at the scene. """
# 拉取 Cosmos 3 NIM 容器 docker run --gpus=all \ -e NGC_API_KEY=$NGC_API_KEY \ -e NIM_MODEL_SIZE=nano \ -p 8000:8000 \ nvcr.io/nim/nvidia/cosmos3-reasoner:latest
# API 调用示例 curl -X POST http://localhost:8000/v1/chat/completions \ -H "Content-Type: application/json" \ -d '{ "model": "cosmos3-nano-reasoner", "messages": [ { "role": "user", "content": [ {"type": "text", "text": "Describe the physical interactions in this video"}, {"type": "video", "video_url": "driver_drowsiness.mp4"} ] } ] }'
七、IMS/DMS 落地应用路线图
7.1 数据合成 Pipeline
graph LR
A[场景定义] --> B[Prompt 编写]
B --> C[Cosmos 3 生成]
C --> D[质量评估<br/>HUE Benchmark]
D -->|通过| E[标注自动化<br/>3D眼球/姿态]
D -->|不通过| B
E --> F[数据集构建]
F --> G[模型训练]
G --> H[部署验证]