Implementation:Tensorflow Tfjs RandomWidth Layer

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

Overview

The RandomWidth layer is a data augmentation preprocessing layer that randomly adjusts the width of images during training. The width is scaled by a random factor drawn uniformly from the range [1 + widthLower, 1 + widthUpper]. The layer supports both bilinear and nearest interpolation methods. It extends BaseRandomLayer for reproducible random number generation.

Code Reference

Source Location

tfjs-layers/src/layers/preprocessing/random_width.ts (GitHub)

Key Imports

import {image, Rank, serialization, Tensor, tidy} from '@tensorflow/tfjs-core';
import {BaseRandomLayerArgs, BaseRandomLayer} from '../../engine/base_random_layer';
import {randomUniform} from '@tensorflow/tfjs-core';

Layer Class

export class RandomWidth extends BaseRandomLayer {
  static override className = 'RandomWidth';
  constructor(args: RandomWidthArgs);
  override getConfig(): serialization.ConfigDict;
  override computeOutputShape(inputShape: Shape | Shape[]): Shape | Shape[];
  override call(inputs: Tensor<Rank.R3> | Tensor<Rank.R4>,
                kwargs: Kwargs): Tensor[] | Tensor;
}

RandomWidthArgs

export interface RandomWidthArgs extends BaseRandomLayerArgs {
  factor: number | [number, number];        // width adjustment factor range
  interpolation?: InterpolationType;        // 'bilinear' (default) or 'nearest'
  seed?: number;
  autoVectorize?: boolean;
}

Factor Interpretation

• Single number (e.g., 0.2): Interpreted as [-0.2, 0.2], so width is scaled by a factor in [0.8, 1.2].
• Tuple (e.g., [-0.1, 0.3]): Width is scaled by a factor in [0.9, 1.3].
• Both bounds must be >= -1.0, and upper bound must be >= lower bound.

I/O Contract

Method	Input	Output
call	3D or 4D image tensor	Tensor with randomly adjusted width, height unchanged
computeOutputShape	Input shape	[imgHeight, -1, numChannels] (width is dynamic)

Usage Example

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

const augmenter = tf.layers.randomWidth({
  factor: 0.2,              // width varies by +/- 20%
  interpolation: 'bilinear',
  seed: 42
});

const images = tf.randomNormal([4, 100, 100, 3]);
const augmented = augmenter.apply(images, {training: true});
// width will be randomly adjusted between 80 and 120 pixels

Related Pages

• Tensorflow_Tfjs_RandomHeight_Layer - Random height augmentation layer (symmetric counterpart)
• Tensorflow_Tfjs_Resizing_Layer - Deterministic image resizing layer
• Tensorflow_Tfjs_CenterCrop_Layer - Center crop preprocessing layer