Implementation:Tensorflow Tfjs CPU ResizeBilinear Kernel

Knowledge Sources	Tensorflow_Tfjs
Domains	Deep_Learning, CPU_Backend
Last Updated	2026-02-10 06:00 GMT

Overview

The ResizeBilinear kernel resizes 4D image tensors to a specified size using bilinear interpolation on the CPU backend. It supports both alignCorners and halfPixelCenters modes, which control how source pixel coordinates are computed during the interpolation.

Code Reference

Source Location

tfjs-backend-cpu/src/kernels/ResizeBilinear.ts (GitHub)

Signature

export function resizeBilinear(args: {
  inputs: ResizeBilinearInputs,
  backend: MathBackendCPU,
  attrs: ResizeBilinearAttrs
}): TensorInfo

Imports

import {KernelConfig, KernelFunc, ResizeBilinear, ResizeBilinearAttrs, ResizeBilinearInputs, TensorInfo, TypedArray, util} from '@tensorflow/tfjs-core';
import {MathBackendCPU} from '../backend_cpu';
import {assertNotComplex} from '../cpu_util';

Kernel Config

export const resizeBilinearConfig: KernelConfig = {
  kernelName: ResizeBilinear,
  backendName: 'cpu',
  kernelFunc: resizeBilinear as unknown as KernelFunc
};

I/O Contract

Inputs

Name	Type	Description
images	TensorInfo	4D input tensor of shape [batch, height, width, channels]

Attributes

Name	Type	Description
size	[number, number]	Target [newHeight, newWidth]
alignCorners	boolean	If true, aligns the corner pixels of input and output
halfPixelCenters	boolean	If true, pixel centers are at half-pixel offsets

Output

Returns a TensorInfo with float32 dtype and shape [batch, newHeight, newWidth, channels].

Algorithm

1. Computes effective input and output sizes based on alignCorners setting.
2. Calculates row and column scaling ratios: effectiveInputSize / effectiveOutputSize.
3. For each output pixel [b, r, c, d]:
   1. Computes the fractional source row and column. When halfPixelCenters is true, applies a 0.5-pixel offset.
   2. Determines the floor and ceil source rows/columns, clamped to valid bounds.
   3. Retrieves the four nearest neighbor values: topLeft, topRight, bottomLeft, bottomRight.
   4. Performs bilinear interpolation: first interpolates horizontally (top and bottom), then vertically between the results.
4. Returns the result as a new tensor via backend.makeTensorInfo.

Usage Example

import * as tf from '@tensorflow/tfjs-core';
import '@tensorflow/tfjs-backend-cpu';

const image = tf.tensor4d([1, 2, 3, 4], [1, 2, 2, 1]);
const resized = tf.image.resizeBilinear(image, [4, 4]);
resized.print();

Related Pages

• Principle:Tensorflow_Tfjs_CPU_Kernel_Registration