Implementation:Tensorflow Tfjs Rescaling Layer

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

Overview

The Rescaling layer is an image preprocessing layer that applies a linear transformation to pixel values: output = input * scale + offset. It is commonly used to normalize pixel values from [0, 255] to [0, 1] or [-1, 1] as the first step in an image processing pipeline. The layer automatically casts inputs to float32 if needed.

Code Reference

Source Location

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

Key Imports

import {LayerArgs, Layer} from '../../engine/topology';
import {serialization, Tensor, mul, add, tidy} from '@tensorflow/tfjs-core';
import {getExactlyOneTensor} from '../../utils/types_utils';
import * as K from '../../backend/tfjs_backend';

Layer Class

export class Rescaling extends Layer {
  static className = 'Rescaling';
  constructor(args: RescalingArgs);
  override getConfig(): serialization.ConfigDict;
  override call(inputs: Tensor | Tensor[], kwargs: Kwargs): Tensor[] | Tensor;
}

RescalingArgs

export interface RescalingArgs extends LayerArgs {
  scale: number;     // multiplicative factor
  offset?: number;   // additive offset (default: 0)
}

I/O Contract

Method	Input	Output
call	Tensor of any shape (cast to float32)	input * scale + offset (same shape, float32)

Usage Example

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

// Normalize pixel values from [0, 255] to [0, 1]
const model = tf.sequential();
model.add(tf.layers.rescaling({scale: 1.0 / 255, inputShape: [224, 224, 3]}));

// Normalize to [-1, 1]
const model2 = tf.sequential();
model2.add(tf.layers.rescaling({
  scale: 1.0 / 127.5,
  offset: -1,
  inputShape: [224, 224, 3]
}));

Related Pages

• Tensorflow_Tfjs_Resizing_Layer - Image resizing preprocessing layer
• Tensorflow_Tfjs_CenterCrop_Layer - Center crop preprocessing layer
• Tensorflow_Tfjs_CategoryEncoding_Layer - Categorical feature preprocessing