3D Generation API

API reference for the 3D Generation module.

ModelGenerator

Main class for generating 3D models from 2D masks.

Constructor

ModelGenerator(
    method: str = 'midas',
    resolution: int = 256,
    device: str = 'cuda'
)

Parameters:

• method (str): Depth estimation method ('midas', 'silhouette')
• resolution (int): Mesh resolution
• device (str): Compute device ('cuda' or 'cpu')

Example:

generator = ModelGenerator(method='midas', resolution=512)

Methods

generate_from_mask()

Generate 3D model from 2D segmentation mask.

def generate_from_mask(
    mask: np.ndarray,
    depth_scale: float = 1.0,
    mesh_simplification: float = 1.0
) -> Mesh3D

Parameters:

• mask (np.ndarray): Binary segmentation mask
• depth_scale (float): Scale factor for depth values
• mesh_simplification (float): Simplification ratio (0-1)

Returns: 3D mesh object

---

save_model()

Save 3D model to file.

def save_model(
    model: Mesh3D,
    output_path: str,
    format: str = 'obj'
) -> None

Parameters:

• model (Mesh3D): 3D mesh to save
• output_path (str): Output file path
• format (str): File format ('obj', 'ply', 'stl', 'gltf')

---

batch_generate()

Generate multiple 3D models in batch.

def batch_generate(
    masks: list,
    num_workers: int = 4
) -> list

Parameters:

• masks (list): List of binary masks
• num_workers (int): Number of parallel workers

Returns: List of 3D mesh objects

---

estimate_depth()

Estimate depth map from mask.

def estimate_depth(mask: np.ndarray) -> np.ndarray

Parameters:

• mask (np.ndarray): Binary segmentation mask

Returns: Depth map as numpy array

---

simplify_mesh()

Reduce mesh polygon count.

def simplify_mesh(
    model: Mesh3D,
    target_faces: int,
    preserve_boundary: bool = True
) -> Mesh3D

Parameters:

• model (Mesh3D): Input mesh
• target_faces (int): Target number of faces
• preserve_boundary (bool): Preserve mesh boundaries

Returns: Simplified mesh

---

smooth_mesh()

Apply smoothing to mesh.

def smooth_mesh(
    model: Mesh3D,
    iterations: int = 10,
    method: str = 'laplacian'
) -> Mesh3D

Parameters:

• model (Mesh3D): Input mesh
• iterations (int): Number of smoothing iterations
• method (str): Smoothing method ('laplacian', 'taubin')

Returns: Smoothed mesh