Module: tfm.vision.layers

Layers package definition.

Classes

class BottleneckBlock: A standard bottleneck block.

class BottleneckBlock3D: Creates a 3D bottleneck block.

class BottleneckResidualInner: Creates a single inner block of a bottleneck.

class BoxSampler: Creates a BoxSampler to sample positive and negative boxes.

class CausalConvMixin: Mixin class to implement CausalConv for tf.keras.layers.Conv layers.

class Conv2D: Conv2D layer supporting CausalConv.

class Conv3D: Conv3D layer supporting CausalConv.

class DepthwiseConv2D: DepthwiseConv2D layer supporting CausalConv.

class DepthwiseSeparableConvBlock: Creates an depthwise separable convolution block with batch normalization.

class DetectionGenerator: Generates the final detected boxes with scores and classes.

class GlobalAveragePool3D: Creates a global average pooling layer with causal mode.

class InvertedBottleneckBlock: An inverted bottleneck block.

class MaskSampler: Samples and creates mask training targets.

class MultilevelDetectionGenerator: Generates detected boxes with scores and classes for one-stage detector.

class MultilevelROIAligner: Performs ROIAlign for the second stage processing.

class MultilevelROIGenerator: Proposes RoIs for the second stage processing.

class PositionalEncoding: Creates a network layer that adds a sinusoidal positional encoding.

class ROISampler: Samples ROIs and assigns targets to the sampled ROIs.

class ResidualBlock: A residual block.

class ResidualInner: Creates a single inner block of a residual.

class ReversibleLayer: Creates a reversible layer.

class Scale: Scales the input by a trainable scalar weight.

class SelfGating: Feature gating as used in S3D-G.

class SpatialAveragePool3D: Creates a global average pooling layer pooling across spatial dimentions.

class SqueezeExcitation: Creates a squeeze and excitation layer.

class StochasticDepth: Creates a stochastic depth layer.

class TemporalSoftmaxPool: Creates a network layer corresponding to temporal softmax pooling.