megengine.functional.nn.roi_pooling

roi_pooling(inp, rois, output_shape, mode='max', scale=1.0)

Applies RoI (Region of Interest) pooling on input feature, as described in Faster RCNN.

See also

• Region of interest pooling explained

• Faster R-CNN

Parameters:

• inp (Tensor) – the input tensor that represents the input feature with (n, c, h, w) shape.

• rois (Tensor) – a tensor represents Regions of Interest with shape (K, 5), which means total K box coordinates in (idx, x1, y1, x2, y2) format where the regions will be taken from. The coordinate including (x1, y1) and (x2, y2) must satisfy 0 <= x1 < x2 and 0 <= y1 < y2. The first column idx should contain the index of the corresponding element in the input batch, i.e. a number in [0, n - 1].

• mode (str) – “max” or “average”, the pooling mode to be used. Default: “max”

• scale (float) – It is a scale that maps output rois feature to input feature. For example, if the output is 224 * 224 image, and the input is a 112 * 112 feature map, then the scale should be set to 0.5. The default value is 1.0

Return type:

Tensor

Returns:

output tensor. (K, C, output_shape[0], output_shape[1]) feature of rois.

Examples

>>> import numpy as np
>>> np.random.seed(42)
>>> inp = Tensor(np.random.randn(1, 1, 128, 128))
>>> rois = Tensor(np.random.random((4, 5)))
>>> y = F.vision.roi_pooling(inp, rois, (2, 2))
>>> y.numpy()[0].round(decimals=4)
array([[[-0.1383, -0.1383],
        [-0.5035, -0.5035]]], dtype=float32)