box_embeddings.modules.pooling.bag_of_boxes

Module Contents

bag_of_boxes_pooler(boxes: box_embeddings.parameterizations.box_tensor.BoxTensor, mask: Optional[torch.BoolTensor] = None, weights: Optional[torch.Tensor] = None, dim: int = 0, keepdim: bool = False) → box_embeddings.parameterizations.box_tensor.BoxTensor
class BagOfBoxesBoxPooler(dim: int = 0, keepdim: bool = False)

Bases: box_embeddings.modules.pooling.pooling.BoxPooler

Pools a box tensor using hard intersection operation

forward(self, box_tensor: box_embeddings.parameterizations.box_tensor.BoxTensor, mask: torch.BoolTensor = None, weights: torch.BoolTensor = None) → box_embeddings.parameterizations.box_tensor.BoxTensor
Parameters

  • box_tensor – Input

  • mask – With shape as box_tensor.box_shape[:-1]. 0 at a position means mask it.

  • weights – With shape as box_tensor.box_shape[:-1].

Returns

Pooled output

Return type

BoxTensor