ConvLSTMLayer class

A Convolutional Long Short-Term Memory (ConvLSTM) layer.

A ConvLSTMLayer is designed for spatiotemporal data, such as a sequence of images (a video). It combines the spatial processing of a Conv2D layer with the temporal memory of an LSTMLayer.

Instead of using matrix multiplications, its internal gates use convolutions. This means its hidden state and cell state are not vectors, but 2D feature maps that preserve spatial information from one timestep to the next.

• Input: A Tensor<Tensor3D> representing the sequence, with a shape of [sequence_length, height, width].
• Output: A Tensor<Matrix> representing the final hidden state, which is a feature map of shape [height, width].

Analogy 🌦️

A ConvLSTM is like a meteorologist watching a weather radar loop. It uses convolutions to see the shape of a storm in each frame and its LSTM logic to track how that shape moves and changes over time to predict its next location.

Example

Layer convLstm = ConvLSTMLayer(16, 3); // 16 hidden filters, 3x3 kernel

Inheritance

• Object
• Layer
• ConvLSTMLayer

Constructors

ConvLSTMLayer(int hiddenFilters, int kernelSize)

Properties

b_c ↔ Tensor<Matrix>

getter/setter pair

b_f ↔ Tensor<Matrix>

getter/setter pair

b_i ↔ Tensor<Matrix>

getter/setter pair

b_o ↔ Tensor<Matrix>

getter/setter pair

hashCode → int

The hash code for this object.

no setter, inherited

hiddenFilters ↔ int

getter/setter pair

K_hc ↔ Tensor<Matrix>

getter/setter pair

K_hf ↔ Tensor<Matrix>

getter/setter pair

K_hi ↔ Tensor<Matrix>

getter/setter pair

K_ho ↔ Tensor<Matrix>

getter/setter pair

K_xc ↔ Tensor<Matrix>

getter/setter pair

K_xf ↔ Tensor<Matrix>

getter/setter pair

K_xi ↔ Tensor<Matrix>

getter/setter pair

K_xo ↔ Tensor<Matrix>

getter/setter pair

kernelSize ↔ int

getter/setter pair

name ↔ String

A user-friendly name for the layer (e.g., 'dense', 'lstm').

getter/setter pair, override-getter

parameters → List<Tensor>

A list of all trainable tensors (weights and biases) in the layer.

no setter, override

runtimeType → Type

A representation of the runtime type of the object.

no setter, inherited

Methods

build(Tensor input) → void

Initializes the 8 kernels and 4 biases for the LSTM gates.

override

call(Tensor input) → Tensor

The public, callable interface for the layer.

inherited

forward(Tensor input) → Tensor<Matrix>

Performs the forward pass by unrolling the LSTM cell through time.

override

noSuchMethod(Invocation invocation) → dynamic

Invoked when a nonexistent method or property is accessed.

inherited

toString() → String

A string representation of this object.

inherited

Operators

operator ==(Object other) → bool

The equality operator.

inherited