chainer.functions.roi_average_align_2d(x, rois, roi_indices, outsize, spatial_scale, sampling_ratio=None)[source]

Spatial Region of Interest (ROI) average align function.

This function acts similarly to roi_average_pooling_2d(), but it computes average of input spatial patch with bilinear interpolation for each channel with the region of interest.

  • x (Variable) – Input variable. The shape is expected to be 4 dimentional: (n: batch, c: channel, h, height, w: width).

  • rois (Variable) – Input roi variable. The shape is expected to be (n: data size, 4), and each datum is set as below: (y_min, x_min, y_max, x_max).

  • roi_indices (Variable) – Input roi variable. The shape is expected to be (n: data size, ).

  • outsize ((int, int) or int) – Expected output size after pooled (height, width). outsize=o and outsize=(o, o) are equivalent.

  • spatial_scale (float) – Scale of the roi is resized.

  • sampling_ratio ((int, int) or int) – Sampling step for the alignment. It must be an integer over \(1\) or None, and the value is automatically decided when None is passed. Use of different ratio in height and width axis is also supported by passing tuple of int as (sampling_ratio_h, sampling_ratio_w). sampling_ratio=s and sampling_ratio=(s, s) are equivalent.


Output variable.

Return type


See the original paper proposing ROIAlign: Mask R-CNN.