torchsparse.nn.functional#

relu(input: SparseTensor, inplace: bool = True) → SparseTensor[source]#

leaky_relu(input: SparseTensor, negative_slope: float = 0.1, inplace: bool = True) → SparseTensor[source]#

build_kernel_map(_coords: Tensor, kernel_size: int | Tuple[int, ...] = 2, stride: int | Tuple[int, ...] = 2, tensor_stride: int | Tuple[int, ...] = 1, mode='hashmap') → Tensor[source]#

conv3d(input: SparseTensor, weight: Tensor, kernel_size: int | List[int] | Tuple[int, ...], bias: Tensor | None = None, stride: int | List[int] | Tuple[int, ...] = 1, dilation: int | Tuple[int, ...] = 1, transposed: bool = False, epsilon: float = 0.0, mm_thresh: int = 0, kmap_mode: str = 'hashmap') → SparseTensor[source]#

spcount(coords: Tensor, num: Tensor) → Tensor[source]#

spcrop(input: SparseTensor, coords_min: Tuple[int, ...] | None = None, coords_max: Tuple[int, ...] | None = None) → SparseTensor[source]#

spdevoxelize(feats: Tensor, coords: Tensor, weights: Tensor) → Tensor[source]#

calc_ti_weights(coords: Tensor, idx_query: Tensor, scale: float = 1) → Tensor[source]#

spdownsample(coords: Tensor, stride: int | Tuple[int, ...] = 2, kernel_size: int | Tuple[int, ...] = 2, tensor_stride: int | Tuple[int, ...] = 1) → Tensor[source]#

sphash(coords: Tensor, offsets: Tensor | None = None) → Tensor[source]#

global_avg_pool(inputs: SparseTensor) → Tensor[source]#

global_max_pool(inputs: SparseTensor) → Tensor[source]#

sphashquery(queries: Tensor, references: Tensor) → Tensor[source]#

spvoxelize(feats: Tensor, coords: Tensor, counts: Tensor) → Tensor[source]#