openvino.runtime.opset1.deformable_psroi_pooling#

openvino.runtime.opset1.deformable_psroi_pooling(feature_maps: Node | int | float | ndarray, coords: Node | int | float | ndarray, output_dim: int, spatial_scale: float, group_size: int = 1, mode: str = 'bilinear_deformable', spatial_bins_x: int = 1, spatial_bins_y: int = 1, trans_std: float = 1.0, part_size: int = 1, offsets: Node | int | float | ndarray | None = None, name: str | None = None) Node#

Return node performing DeformablePSROIPooling operation.

DeformablePSROIPooling computes position-sensitive pooling on regions of interest specified by input.

Parameters:
  • feature_maps – 4D tensor with feature maps.

  • coords – 2D tensor describing box consisting of tuples: [batch_id, x_1, y_1, x_2, y_2].

  • output_dim – A pooled output channel number.

  • spatial_scale – A multiplicative spatial scale factor to translate ROI.

  • group_size – The number of groups to encode position-sensitive score.

  • mode – Specifies mode for pooling. Range of values: [‘bilinear_deformable’].

  • spatial_bins_x – Specifies numbers of bins to divide the input feature maps over width.

  • spatial_bins_y – Specifies numbers of bins to divide the input feature maps over height.

  • trans_std – The value that all transformation (offset) values are multiplied with.

  • part_size – The number of parts the output tensor spatial dimensions are divided into.

  • offsets – Optional node. 4D input blob with transformation values (offsets).

  • name – The optional new name for output node.

Returns:

New node performing DeformablePSROIPooling operation.