dgs.models.metric.metric.CosineSimilarityMetric

class dgs.models.metric.metric.CosineSimilarityMetric(*args: Any, **kwargs: Any)[source]

Class to compute the cosine similarity between two matrices.

Notes

The cosine similarity is defined as:

\[\text{cosine similarity} = S_C(A,B) = \frac{\mathbf{A} \cdot \mathbf{B}}{\|\mathbf{A}\| \|\mathbf{B}\|}\]

Methods

__init__(*args, **kwargs) → None[source]
forward(input1: torch.Tensor, input2: torch.Tensor) → torch.Tensor[source]

Due to the sheer size of the PoseTrack21 dataset, CosineSimilarity() does not work due to memory issues. See this issue <https://github.com/pytorch/pytorch/issues/104564#issuecomment-1625348908>_ for more details.

Parameters:

  • input1 – tensor of shape [a x E]

  • input2 – tensor of shape [b x E]

Returns:

tensor of shape [a x b] containing the distances.

References

https://github.com/pytorch/pytorch/issues/104564#issuecomment-1625348908