The WiML Symposium 2023 returns December 7! Join the livestream at 9:30 AM PST Register

tfm.vision.layers.MultilevelROIAligner

Performs ROIAlign for the second stage processing.

tfm.vision.layers.MultilevelROIAligner(
    crop_size: int = 7, sample_offset: float = 0.5, **kwargs
)

Args
`crop_size`	An `int` of the output size of the cropped features.
`sample_offset`	A `float` in [0, 1] of the subpixel sample offset.
`**kwargs`	Additional keyword arguments passed to Layer.

Methods

call(
    features: Mapping[str, tf.Tensor], boxes: tf.Tensor, training: bool = None
)

Generates ROIs.

Args
`features`	A dictionary with key as pyramid level and value as features. The features are in shape of [batch_size, height_l, width_l, num_filters].
`boxes`	A 3-D `tf.Tensor` of shape [batch_size, num_boxes, 4]. Each row represents a box with [y1, x1, y2, x2] in un-normalized coordinates. from grid point.
`training`	A `bool` of whether it is in training mode.

Returns
A 5-D `tf.Tensor` representing feature crop of shape [batch_size, num_boxes, crop_size, crop_size, num_filters].