Implementation:Tensorflow Tfjs Padding Layers

Knowledge Sources	Tensorflow_Tfjs TensorFlow.js API
Domains	Deep_Learning, Layers_API, Convolutional
Last Updated	2026-02-10 06:00 GMT

Overview

This module implements padding layers and utility functions for TensorFlow.js Layers. It provides temporalPadding for 3D tensors (sequence data), spatial2dPadding for 4D tensors (image data), and the ZeroPadding2D layer class that adds rows and columns of zeros to the spatial dimensions of a 4D image tensor. These are ported from Keras convolutional.py but placed in a separate file for clarity.

Code Reference

Source Location

tfjs-layers/src/layers/padding.ts (GitHub)

Key Imports

import * as tfc from '@tensorflow/tfjs-core';
import {serialization, Tensor, tidy} from '@tensorflow/tfjs-core';
import {imageDataFormat} from '../backend/common';
import {InputSpec, Layer, LayerArgs} from '../engine/topology';

Utility Functions

temporalPadding

Pads the middle (temporal) dimension of a 3D tensor with zeros.

export function temporalPadding(x: Tensor, padding?: [number, number]): Tensor

Default padding is [1, 1]. Input must be rank 3.

spatial2dPadding

Pads the 2nd and 3rd spatial dimensions of a 4D tensor, respecting channelsFirst or channelsLast data format.

export function spatial2dPadding(
    x: Tensor,
    padding?: [[number, number], [number, number]],
    dataFormat?: DataFormat): Tensor

Default padding is [[1, 1], [1, 1]].

Layer Class

ZeroPadding2D

export class ZeroPadding2D extends Layer {
  static className = 'ZeroPadding2D';
  readonly dataFormat: DataFormat;
  readonly padding: [[number, number], [number, number]];
  constructor(args?: ZeroPadding2DLayerArgs);
  override computeOutputShape(inputShape: Shape | Shape[]): Shape | Shape[];
  override call(inputs: Tensor | Tensor[], kwargs: Kwargs): Tensor | Tensor[];
  override getConfig(): serialization.ConfigDict;
}

ZeroPadding2DLayerArgs

The padding parameter accepts three formats:

• Integer: Symmetric padding applied to both height and width.
• [number, number]: Symmetric padding [heightPad, widthPad].
• [[number, number], [number, number]]: Explicit [[topPad, bottomPad], [leftPad, rightPad]].

I/O Contract

Operation	Input	Output
temporalPadding	3D Tensor [batch, time, features]	3D Tensor with padded time dimension
spatial2dPadding	4D Tensor	4D Tensor with padded spatial dimensions
ZeroPadding2D.call	4D Tensor [batch, h, w, c] (or channelsFirst)	4D Tensor with zero-padded height and width
computeOutputShape	Input shape array	Output shape with increased spatial dims

Usage Example

import * as tf from '@tensorflow/tfjs';

const model = tf.sequential();
model.add(tf.layers.zeroPadding2d({
  padding: [[1, 1], [2, 2]],  // 1px top/bottom, 2px left/right
  inputShape: [28, 28, 1]
}));
// Output shape: [null, 30, 32, 1]

Related Pages

• Tensorflow_Tfjs_DepthwiseConv2D_Layer - Depthwise convolution layer often used after padding
• Tensorflow_Tfjs_Resizing_Layer - Image resizing preprocessing layer