Class SAM3MultiObjectSegmenter

preprocessing/sam3_segmenter.py:13–170

SAM3 多物体分割器

class SAM3MultiObjectSegmenter:
    """SAM3 多物体分割器"""
    
    def __init__(self, checkpoint_path: Path = None, confidence_threshold: float = 0.1):
        """
        初始化 SAM3 模型
        
        Args:
            checkpoint_path: SAM3 checkpoint 路径
            confidence_threshold: 置信度阈值
        """
        if checkpoint_path is None:
            checkpoint_path = Path("/mnt/workspace/users/lbc/sam3/checkpoints/sam3.pt")
        
        self.checkpoint_path = checkpoint_path
        self.confidence_threshold = confidence_threshold
        
        # 添加 sam3 路径
        sam3_path = Path("/mnt/workspace/users/lbc/sam3")
        if sam3_path not in [Path(p) for p in sys.path]:
            sys.path.insert(0, str(sam3_path))
        
        # 导入 SAM3
        from sam3.model_builder import build_sam3_image_model
        from sam3.model.sam3_image_processor import Sam3Processor
        
        # 加载模型
        logger.info(f"Loading SAM3 model from: {checkpoint_path}")
        model = build_sam3_image_model(
            checkpoint_path=str(checkpoint_path),
            load_from_HF=False
        )
        self.processor = Sam3Processor(model, confidence_threshold=confidence_threshold)
        logger.success("✓ SAM3 model loaded")
    
    def segment_object_multiview(
        self,
        images_dir: Path,
        object_name: str,
        text_prompt: str,
        output_dir: Path,
    ) -> Dict:
        """
        对多个视角的图像分割同一个物体
        
        Args:
            images_dir: 图像目录
            object_name: 物体名称
            text_prompt: SAM3 文本提示词
            output_dir: 输出目录（将创建 object_name/ 子目录）
        
        Returns:
            Dict with status and info
        """
        logger.info(f"\n[Segmenting: {object_name}]")
        logger.info(f"  Prompt: '{text_prompt}'")
        
        # 获取所有图像（使用自然数字排序，确保 2.png 排在 10.png 前面）