Download node_modules/three/examples/jsm/shaders/ConvolutionShader.js from SimWorld-AI/industrial-harbor-viewer: direct link, hf CLI and curl.
- Browser
- Download file 1.81 kB
-
https://huggingface.co/spaces/SimWorld-AI/industrial-harbor-viewer/resolve/main/node_modules/three/examples/jsm/shaders/ConvolutionShader.js
- Command line
-
hf download hf://spaces/SimWorld-AI/industrial-harbor-viewer/node_modules/three/examples/jsm/shaders/ConvolutionShader.js
-
curl -L -o ConvolutionShader.js https://huggingface.co/spaces/SimWorld-AI/industrial-harbor-viewer/resolve/main/node_modules/three/examples/jsm/shaders/ConvolutionShader.js
1.81 kB
| import { | |
| Vector2 | |
| } from 'three'; | |
| /** | |
| * Convolution shader | |
| * ported from o3d sample to WebGL / GLSL | |
| */ | |
| const ConvolutionShader = { | |
| name: 'ConvolutionShader', | |
| defines: { | |
| 'KERNEL_SIZE_FLOAT': '25.0', | |
| 'KERNEL_SIZE_INT': '25' | |
| }, | |
| uniforms: { | |
| 'tDiffuse': { value: null }, | |
| 'uImageIncrement': { value: new Vector2( 0.001953125, 0.0 ) }, | |
| 'cKernel': { value: [] } | |
| }, | |
| vertexShader: /* glsl */` | |
| uniform vec2 uImageIncrement; | |
| varying vec2 vUv; | |
| void main() { | |
| vUv = uv - ( ( KERNEL_SIZE_FLOAT - 1.0 ) / 2.0 ) * uImageIncrement; | |
| gl_Position = projectionMatrix * modelViewMatrix * vec4( position, 1.0 ); | |
| }`, | |
| fragmentShader: /* glsl */` | |
| uniform float cKernel[ KERNEL_SIZE_INT ]; | |
| uniform sampler2D tDiffuse; | |
| uniform vec2 uImageIncrement; | |
| varying vec2 vUv; | |
| void main() { | |
| vec2 imageCoord = vUv; | |
| vec4 sum = vec4( 0.0, 0.0, 0.0, 0.0 ); | |
| for( int i = 0; i < KERNEL_SIZE_INT; i ++ ) { | |
| sum += texture2D( tDiffuse, imageCoord ) * cKernel[ i ]; | |
| imageCoord += uImageIncrement; | |
| } | |
| gl_FragColor = sum; | |
| }`, | |
| buildKernel: function ( sigma ) { | |
| // We lop off the sqrt(2 * pi) * sigma term, since we're going to normalize anyway. | |
| const kMaxKernelSize = 25; | |
| let kernelSize = 2 * Math.ceil( sigma * 3.0 ) + 1; | |
| if ( kernelSize > kMaxKernelSize ) kernelSize = kMaxKernelSize; | |
| const halfWidth = ( kernelSize - 1 ) * 0.5; | |
| const values = new Array( kernelSize ); | |
| let sum = 0.0; | |
| for ( let i = 0; i < kernelSize; ++ i ) { | |
| values[ i ] = gauss( i - halfWidth, sigma ); | |
| sum += values[ i ]; | |
| } | |
| // normalize the kernel | |
| for ( let i = 0; i < kernelSize; ++ i ) values[ i ] /= sum; | |
| return values; | |
| } | |
| }; | |
| function gauss( x, sigma ) { | |
| return Math.exp( - ( x * x ) / ( 2.0 * sigma * sigma ) ); | |
| } | |
| export { ConvolutionShader }; | |