tfm.vision.heads.MaskScoring

Creates a mask scoring layer.

tfm.vision.heads.MaskScoring(
    num_classes: int,
    fc_input_size: List[int],
    num_convs: int = 3,
    num_filters: int = 256,
    use_depthwise_convolution: bool = False,
    fc_dims: int = 1024,
    num_fcs: int = 2,
    activation: str = 'relu',
    use_sync_bn: bool = False,
    norm_momentum: float = 0.99,
    norm_epsilon: float = 0.001,
    kernel_regularizer: Optional[tf.keras.regularizers.Regularizer] = None,
    bias_regularizer: Optional[tf.keras.regularizers.Regularizer] = None,
    **kwargs
)

This implements mask scoring layer from the paper:

Zhaojin Huang, Lichao Huang, Yongchao Gong, Chang Huang, Xinggang Wang. Mask Scoring R-CNN. (https://arxiv.org/pdf/1903.00241.pdf)

Args
num_classes	An int for number of classes.
fc_input_size	A List of int for the input size of the fully connected layers.
num_convs	Anint for number of conv layers.
num_filters	An int for the number of filters for conv layers.
use_depthwise_convolution	A bool, whether or not using depthwise convs.
fc_dims	An int number of filters for each fully connected layers.
num_fcs	An int for number of fully connected layers.
activation	A str name of the activation function.
use_sync_bn	A bool, whether or not to use sync batch normalization.
norm_momentum	A float for the momentum in BatchNorm. Defaults to 0.99.
norm_epsilon	A float for the epsilon value in BatchNorm. Defaults to 0.001.
kernel_regularizer	A tf.keras.regularizers.Regularizer object for Conv2D. Default is None.
bias_regularizer	A tf.keras.regularizers.Regularizer object for Conv2D.
**kwargs	Additional keyword arguments to be passed.

Methods

call

View source

call(
    inputs: tf.Tensor, training: bool = None
)

Forward pass mask scoring head.

Args
inputs	A tf.Tensor of the shape [batch_size, width, size, num_classes], representing the segmentation logits.
training	a bool indicating whether it is in training mode.

Returns
mask_scores	A tf.Tensor of predicted mask scores [batch_size, num_classes].