First-Impression-llm-version/example_usage.py

"""
使用示例：展示如何使用不同模态配置进行人格分析
"""
import os
from config import Config
from model_llm import PersonalityAnalyzer
from data_preprocessing import DataPreprocessor


def example_vision_only():
    """示例1: 仅使用视觉模态"""
    print("=" * 50)
    print("示例1: 仅视觉模态")
    print("=" * 50)
    
    # 配置模态
    Config.set_modalities(text=False, vision=True, audio=False)
    
    # 初始化
    analyzer = PersonalityAnalyzer()
    preprocessor = DataPreprocessor()
    
    # 假设有一个视频文件
    video_path = "../First-Impression/data/example.mp4"
    
    # 提取图像
    images = preprocessor.extract_images_from_video(
        video_path,
        "./temp_frames",
        max_frames=10
    )
    
    # 预测
    result = analyzer.predict(images=images)
    
    print("\n预测结果:")
    for trait, score in result["scores"].items():
        print(f"  {trait}: {score:.3f}")


def example_text_only():
    """示例2: 仅使用文本模态"""
    print("=" * 50)
    print("示例2: 仅文本模态")
    print("=" * 50)
    
    # 配置模态
    Config.set_modalities(text=True, vision=False, audio=False)
    
    # 初始化
    analyzer = PersonalityAnalyzer()
    
    # 文本描述
    text_description = """
    这个人在视频中表现出以下特征：
    - 说话声音较大，语速较快
    - 经常微笑，表情丰富
    - 手势较多，身体语言活跃
    - 与镜头有良好的眼神交流
    """
    
    # 预测
    result = analyzer.predict(text=text_description)
    
    print("\n预测结果:")
    for trait, score in result["scores"].items():
        print(f"  {trait}: {score:.3f}")


def example_multimodal():
    """示例3: 多模态组合"""
    print("=" * 50)
    print("示例3: 多模态组合（文本+视觉+音频）")
    print("=" * 50)
    
    # 配置模态
    Config.set_modalities(text=True, vision=True, audio=True)
    
    # 初始化
    analyzer = PersonalityAnalyzer()
    preprocessor = DataPreprocessor()
    
    # 假设有一个视频文件
    video_path = "../First-Impression/data/example.mp4"
    
    # 提取所有模态数据
    images = preprocessor.extract_images_from_video(
        video_path,
        "./temp_frames",
        max_frames=10
    )
    
    audio_path = preprocessor.extract_audio_from_video(
        video_path,
        "./temp_audio"
    )
    
    text_path = preprocessor.extract_text_from_video(
        video_path,
        "./temp_text"
    )
    
    # 读取文本
    text = None
    if text_path and os.path.exists(text_path):
        with open(text_path, "r", encoding="utf-8") as f:
            text = f.read()
    
    # 预测
    result = analyzer.predict(
        text=text,
        images=images,
        audio_path=audio_path
    )
    
    print("\n预测结果:")
    for trait, score in result["scores"].items():
        print(f"  {trait}: {score:.3f}")
    
    if result.get("reasoning"):
        print(f"\n分析理由:\n{result['reasoning']}")


if __name__ == "__main__":
    import os
    
    print("注意: 这些示例需要:")
    print("1. 已安装所有依赖")
    print("2. 已下载 QwenVL2.5-7B 模型")
    print("3. 有可用的视频文件")
    print("\n取消注释下面的函数调用来运行示例:\n")
    
    # 取消注释以运行示例
    # example_vision_only()
    # example_text_only()
    # example_multimodal()