PoLArMAE

class PoLArMAE(arch: Literal['vit_tiny', 'vit_small', 'vit_base'] = 'vit_small', num_channels: int = 4, voxel_size: float = 5, masking_ratio: float = 0.6, masking_type: Literal['rand'] = 'rand', decoder_arch: Literal['vit_tiny', 'vit_small', 'vit_base'] = 'vit_small', decoder_kwargs: dict | None = None, mae_prediction: Literal['full', 'pos'] = 'full', loss_weights: Dict[str, float] | None = None, transformer_kwargs: dict | None = None, tokenizer_kwargs: dict | None = None, decoder_use_kv: bool = True, patch_encoder_kwargs: dict | None = None)

Bases: PointModel

Self-contained PoLAr-MAE for pimm’s DefaultTrainer.

PoLArMAE.forward(data_dict)