tf.keras.layers.TimeDistributed

This wrapper allows to apply a layer to every temporal slice of an input.

Inherits From: Wrapper, Layer, Module

Every input should be at least 3D, and the dimension of index one of the first input will be considered to be the temporal dimension.

Consider a batch of 32 video samples, where each sample is a 128x128 RGB image with channels_last data format, across 10 timesteps. The batch input shape is (32, 10, 128, 128, 3).

You can then use TimeDistributed to apply the same Conv2D layer to each of the 10 timesteps, independently:

inputs = tf.keras.Input(shape=(10, 128, 128, 3))
conv_2d_layer = tf.keras.layers.Conv2D(64, (3, 3))
outputs = tf.keras.layers.TimeDistributed(conv_2d_layer)(inputs)
outputs.shape
TensorShape([None, 10, 126, 126, 64])

Because TimeDistributed applies the same instance of Conv2D to each of the timestamps, the same set of weights are used at each timestamp.

layer a tf.keras.layers.Layer instance.

inputs Input tensor of shape (batch, time, ...) or nested tensors, and each of which has shape (batch, time, ...).
training Python boolean indicating whether the layer should behave in training mode or in inference mode. This argument is passed to the wrapped layer (only if the layer supports this argument).
mask Binary tensor of shape (samples, timesteps) indicating whether a given timestep should be masked. This argument is passed to the wrapped layer (only if the layer supports this argument).

ValueError If not initialized with a tf.keras.layers.Layer instance.