PoLArMAESemSeg

class PoLArMAESemSeg(num_classes: int = 5, arch: Literal['vit_tiny', 'vit_small', 'vit_base'] = 'vit_small', voxel_size: float = 5, num_channels: int = 4, seg_head_fetch_layers: List[int] = [3, 7, 11], seg_head_combination_method: Literal['mean', 'concat'] = 'mean', seg_head_dim: int = 384, seg_head_dropout: float = 0.5, freeze_encoder: bool = False, apply_encoder_postnorm: bool = True, condition_global_features: bool = False, upsampling_k: int = 5, upsampling_dim: int | None = None, center: List[float] = [384.0, 384.0, 384.0], scale: float = 0.0015035163260146505, criteria=None, transformer_kwargs: dict | None = None, tokenizer_kwargs: dict | None = None)

Bases: PointModel

PoLAr-MAE semantic segmentation for pimm’s DefaultTrainer.

PoLArMAESemSeg.forward(data_dict)