1. 技术文章
  2. WebGL 3d usage for depth sorting 2d objects

WebGL 3d usage for depth sorting 2d objects

由 ぃ、小莉子 提交于 2019-12-14 03:59:24

问题


This question has a strong relation with my other question: Isometric rendering without tiles, is that goal reachable?

I want to depth sort objects in an isometric world (html5 canvas). The world is not tiled, so every item in the world can be placed on each x, y, z coordinate. Since it's not a tiled world, depth sorting is hard to do. I even want that if items intersect, that the visible parts are drawn as if it were intersecting parts in a fully 3d world. As people answered in my other question, this can be done by representing each 2d image as a 3d model. I want to go on with the solution given in the following comment on that question:

You don't have to work in 3D when you use webGL. WebGL draws polygons and is very quick at drawing 2D images as 4 verts making a small fan of triangles. You can still use the zbuffer and set corners (verts) to the z distance. Most of the 2D game libraries use webGL to render 2D and fallback to canvas if webGL is not there. There is also a webGL implementation of the canvas API on github that you could modify to meet your needs. (comment link)

So, you could see the 'logic' as 3d models. The z-buffer of webGL provides correct rendering. The rendering pixels itself are pixels of the 2d images. But I don't know how to do this. Could someone further explain how to get this done? I read a lot of information, but that's all about real 3d.


回答1:


Could could use depth sprites as you pointed out in your other question (ps, you really should put those images in this question)

To use depth sprites you need to enable the EXT_frag_depth extension if it exists. Then you can write to gl_fragDepthEXT in your fragment shader. Making depth sprites sounds like more work to me than making 3D models.

In that case you just load 2 textures per sprite, one for color, one for depth and then do something like

 #extension GL_EXT_frag_depth : require

 varying vec2 texcoord;

 uniform sampler2D colorTexture;
 uniform sampler2D depthTexture;
 uniform float depthScale;
 uniform float depthOffset;

 void main() {
   vec4 color = texture2D(colorTexture, texcoord);

   // don't draw if transparent
   if (color.a <= 0.01) {
     discard;
   }

   gl_FragColor = color;

   float depth = texture2D(depthTexture, texcoord).r;
   gl_FragDepthEXT = depthOffset - depth * depthScale;
 }

You'd set depthOffset and depthScale to something like 

 var yTemp = yPosOfSpriteInPixelsFromTopOfScreen + tallestSpriteHeight;
 var depthOffset = 1. - yTemp / 65536;
 var depthScale = 1 / 256;

That assumes each value in the depth texture is less per depth change.

As for how to draw in 2D in WebGL see this article.

Here's an example that seems to work. I generated the image because I'm too lazy to draw it in photoshop. Manually drawing depth values is pretty tedious. It assumes the furthest pixel in the image of depth values of 1, the next closest pixels have a depth value of 2, etc.

In other words if you had a small 3x3 isometric cube the depth values would be something like

+---+---+---+---+---+---+---+---+---+---+
|   |   |   |   | 1 | 1 |   |   |   |   |
+---+---+---+---+---+---+---+---+---+---+
|   |   | 2 | 2 | 2 | 2 | 2 | 2 |   |   |
+---+---+---+---+---+---+---+---+---+---+
| 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 |
+---+---+---+---+---+---+---+---+---+---+
| 3 | 3 | 4 | 4 | 4 | 4 | 4 | 4 | 3 | 3 |
+---+---+---+---+---+---+---+---+---+---+
| 3 | 3 | 4 | 4 | 5 | 5 | 4 | 4 | 3 | 3 |
+---+---+---+---+---+---+---+---+---+---+
| 3 | 3 | 4 | 4 | 5 | 5 | 4 | 4 | 3 | 3 |
+---+---+---+---+---+---+---+---+---+---+
| 3 | 3 | 4 | 4 | 5 | 5 | 4 | 4 | 3 | 3 |
+---+---+---+---+---+---+---+---+---+---+
|   |   | 4 | 4 | 5 | 5 | 4 | 4 |   |   |
+---+---+---+---+---+---+---+---+---+---+
|   |   |   |   | 5 | 5 |   |   |   |   |
+---+---+---+---+---+---+---+---+---+---+


function makeDepthColor(depth) {
  return "rgb(" + depth + "," + depth + "," + depth + ")";
}

function makeSprite(ctx, depth) {
  // make an image (these would be made in photoshop ro
  // some other paint program but that's too much work for me
  ctx.canvas.width = 64;
  ctx.canvas.height = 64;
  for (y = 0; y <= 32; ++y) {
    var halfWidth = (y < 16 ? 1 + y : 33 - y) * 2;
    var width = halfWidth * 2;
    var cy = (16 - y);
    var cw = Math.max(0, 12 - Math.abs(cy) * 2) | 0;
    for (var x = 0; x < width; ++x) {
      var cx = x - halfWidth;
      var inCenter = Math.abs(cy) < 6 && Math.abs(cx) <= cw;
      var onEdge = x < 2 || x >= width - 2 || (inCenter && (Math.abs(cx / 2) | 0) === (cw / 2 | 0));
      var height = onEdge ? 12 : (inCenter ? 30 : 10);
      var color = inCenter ? (cx < 0 ? "#F44" : "#F66") : (cx < 0 ? "#44F" : "#66F");
      ctx.fillStyle = depth ? makeDepthColor(y + 1) : color;
      var xx = 32 - halfWidth + x;
      var yy = y;
      ctx.fillRect(xx, yy + 32 - height, 1, height);
      if (!depth) {
        ctx.fillStyle = onEdge ? "black" : "#CCF";
        ctx.fillRect(xx, yy + 32 - height, 1, 1);
      }
    }
  }
}

function main() {
  var m4 = twgl.m4;
  var gl = document.querySelector("canvas").getContext(
    "webgl", {preserveDrawingBuffer: true});
  var ext = gl.getExtension("EXT_frag_depth");
  if (!ext) {
    alert("need EXT_frag_depth");
    return;
  }

  var vs = `
    attribute vec4 position;
    attribute vec2 texcoord;

    varying vec2 v_texcoord;

    uniform mat4 u_matrix;
    uniform mat4 u_textureMatrix;

    void main() {
      v_texcoord = (u_textureMatrix * vec4(texcoord, 0, 1)).xy;
      gl_Position = u_matrix * position;
    }
  `;

  var fs = `
    #extension GL_EXT_frag_depth : require

    precision mediump float;

    varying vec2 v_texcoord;

    uniform sampler2D u_colorTexture;
    uniform sampler2D u_depthTexture;
    uniform float u_depthScale;
    uniform float u_depthOffset;

    void main() {
      vec4 color = texture2D(u_colorTexture, v_texcoord);
      if (color.a < 0.01) {
        discard;
      }

      float depth = texture2D(u_depthTexture, v_texcoord).r;
      gl_FragDepthEXT = u_depthOffset - depth * u_depthScale;
      gl_FragColor = color;
    }
  `;

  var programInfo = twgl.createProgramInfo(gl, [vs, fs]);
  var quadBufferInfo = twgl.createBufferInfoFromArrays(gl, {
    position: {
      numComponents: 2,
      data: [
        0, 0,
        0, 1,
        1, 0,
        1, 0,
        0, 1,
        1, 1,
      ],
    },
    texcoord: [
      0, 0,
      0, 1,
      1, 0,
      1, 0,
      0, 1,
      1, 1,
    ],
  });

  var ctx = document.createElement("canvas").getContext("2d");

  // make the color texture
  makeSprite(ctx, false);
  var colorTexture = twgl.createTexture(gl, {
    src: ctx.canvas,
    min: gl.NEAREST,
    mag: gl.NEAREST,
  });

  // make the depth texture
  makeSprite(ctx, true);
  var depthTexture = twgl.createTexture(gl, {
    src: ctx.canvas,
    format: gl.LUMINANCE,  // because depth is only 1 channel
    min: gl.NEAREST,
    mag: gl.NEAREST,
  });

  function drawDepthImage(
      colorTex, depthTex, texWidth, texHeight,
      x, y, z) {
    var dstY = y + z;
    var dstX = x;
    var dstWidth = texWidth;
    var dstHeight = texHeight;

    var srcX = 0;
    var srcY = 0;
    var srcWidth = texWidth;
    var srcHeight = texHeight;

    gl.useProgram(programInfo.program);

    twgl.setBuffersAndAttributes(gl, programInfo, quadBufferInfo);

    // this matirx will convert from pixels to clip space
    var matrix = m4.ortho(0, gl.canvas.width, gl.canvas.height, 0, -1, 1);

    // this matrix will translate our quad to dstX, dstY
    matrix = m4.translate(matrix, [dstX, dstY, 0]);

    // this matrix will scale our 1 unit quad
    // from 1 unit to texWidth, texHeight units
    matrix = m4.scale(matrix, [dstWidth, dstHeight, 1]);

    // just like a 2d projection matrix except in texture space (0 to 1)
    // instead of clip space. This matrix puts us in pixel space.
    var texMatrix = m4.scaling([1 / texWidth, 1 / texHeight, 1]);

    // because were in pixel space
    // the scale and translation are now in pixels
    var texMatrix = m4.translate(texMatrix, [srcX, srcY, 0]);
    var texMatrix = m4.scale(texMatrix, [srcWidth, srcHeight, 1]);

    twgl.setUniforms(programInfo, {
      u_colorTexture: colorTex,
      u_depthTexture: depthTex,
      u_matrix: matrix,
      u_textureMatrix: texMatrix,
      u_depthOffset: 1 - (dstY - z) / 65536,
      u_depthScale: 1 / 256,
    });

    twgl.drawBufferInfo(gl, quadBufferInfo);
  }

  // test render
  gl.enable(gl.DEPTH_TEST);

  var texWidth = 64;
  var texHeight = 64;

  // z is how much above/below ground
  function draw(x, y, z) {
    drawDepthImage(colorTexture, depthTexture, texWidth, texHeight , x, y, z);
  }

  draw(  0, 0, 0);  // draw on left

  draw(100, 0, 0);  // draw near center
  draw(113, 0, 0);  // draw overlapping

  draw(200, 0, 0);  // draw on right
  draw(200, 8, 0);  // draw on more forward

  draw(0, 60,  0);  // draw on left
  draw(0, 60, 10);  // draw on below

  draw(100, 60,  0);  // draw near center
  draw(100, 60, 20);  // draw below

  draw(200, 60, 20);  // draw on right
  draw(200, 60,  0);  // draw above
}

main();
<script src="https://twgljs.org/dist/2.x/twgl-full.min.js"></script>
<canvas></canvas>

The top left is what the image looks like. The top middle is 2 images drawn side by side. The top right is 2 images drawn one further down in y (x, y is the iso-plane). The bottom left is two images one drawn below the other (below the plane). The bottom middle is the same thing just separated more. The bottom right is the same thing except drawn in the opposite order (just to check it works)

To save memory you could put the depth value in the alpha channel of the color texture. If it's 0 discard.

Unfortunately according to webglstats.com only 75% of desktops and 0% of phones support EXT_frag_depth. Although WebGL2 requires support for gl_FragDepth and AFAIK most phones support OpenGL ES 3.0 on which WebGL2 is based so in another couple of months most Android phones and most PCs will be getting WebGL2. iOS on the other hand, as usual, Apple is secret about when they will ship WebGL2 on iOS. It's pretty clear they never plan to ship WebGL2 based on the fact that there hasn't been a single commit to WebKit for WebGL2 in over 2 years.

For systems that don't support WebGL2 or EXT_frag_depth on WebGL1 you could simulate EXT_frag_depth using vertex shaders. You'd pass the depth texture to a vertex shader and draw with gl.POINTS, one point per pixel. That way you can choose the depth of each point.

It would work but it might end up being pretty slow. Possibly slower than just doing it in JavaScript directly writing to an array and using Canvas2DRenderingContext.putImageData

Here's an example

function makeDepthColor(depth) {
  return "rgb(" + depth + "," + depth + "," + depth + ")";
}

function makeSprite(ctx, depth) {
  // make an image (these would be made in photoshop ro
  // some other paint program but that's too much work for me
  ctx.canvas.width = 64;
  ctx.canvas.height = 64;
  for (y = 0; y <= 32; ++y) {
    var halfWidth = (y < 16 ? 1 + y : 33 - y) * 2;
    var width = halfWidth * 2;
    var cy = (16 - y);
    var cw = Math.max(0, 12 - Math.abs(cy) * 2) | 0;
    for (var x = 0; x < width; ++x) {
      var cx = x - halfWidth;
      var inCenter = Math.abs(cy) < 6 && Math.abs(cx) <= cw;
      var onEdge = x < 2 || x >= width - 2 || (inCenter && (Math.abs(cx / 2) | 0) === (cw / 2 | 0));
      var height = onEdge ? 12 : (inCenter ? 30 : 10);
      var color = inCenter ? (cx < 0 ? "#F44" : "#F66") : (cx < 0 ? "#44F" : "#66F");
      ctx.fillStyle = depth ? makeDepthColor(y + 1) : color;
      var xx = 32 - halfWidth + x;
      var yy = y;
      ctx.fillRect(xx, yy + 32 - height, 1, height);
      if (!depth) {
        ctx.fillStyle = onEdge ? "black" : "#CCF";
        ctx.fillRect(xx, yy + 32 - height, 1, 1);
      }
    }
  }
}

function main() {
  var m4 = twgl.m4;
  var gl = document.querySelector("canvas").getContext(
    "webgl", {preserveDrawingBuffer: true});
  var numVertexTextures = gl.getParameter(gl.MAX_VERTEX_TEXTURE_IMAGE_UNITS);
  if (numVertexTextures < 2) {
    alert("GPU doesn't support textures in vertex shaders");
    return;
  }

  var vs = `
    attribute float count;

    uniform vec2 u_dstSize;
    uniform mat4 u_matrix;
    uniform mat4 u_textureMatrix;
    uniform sampler2D u_colorTexture;
    uniform sampler2D u_depthTexture;
    uniform float u_depthScale;
    uniform float u_depthOffset;

    varying vec4 v_color;

    void main() {
      float px = mod(count, u_dstSize.x);
      float py = floor(count / u_dstSize.x);

      vec4 position = vec4((vec2(px, py) + 0.5) / u_dstSize, 0, 1);
      vec2 texcoord = (u_textureMatrix * position).xy;

      float depth = texture2D(u_depthTexture, texcoord).r;

      gl_Position = u_matrix * position;
      gl_Position.z = u_depthOffset - depth * u_depthScale;
      v_color = texture2D(u_colorTexture, texcoord);
    }
  `;

  var fs = `
    precision mediump float;

    varying vec4 v_color;

    void main() {
      if (v_color.a < 0.01) {
        discard;
      }
      gl_FragColor = v_color;
    }
  `;
  
  // make a count
  var maxImageWidth = 256;
  var maxImageHeight = 256;
  var maxPixelsInImage = maxImageWidth * maxImageHeight
  var count = new Float32Array(maxPixelsInImage);
  for (var ii = 0; ii < count.length; ++ii) {
    count[ii] = ii;
  }

  var programInfo = twgl.createProgramInfo(gl, [vs, fs]);
  var quadBufferInfo = twgl.createBufferInfoFromArrays(gl, {
    count: { numComponents: 1, data: count, }
  });

  var ctx = document.createElement("canvas").getContext("2d");

  // make the color texture
  makeSprite(ctx, false);
  var colorTexture = twgl.createTexture(gl, {
    src: ctx.canvas,
    min: gl.NEAREST,
    mag: gl.NEAREST,
  });

  // make the depth texture
  makeSprite(ctx, true);
  var depthTexture = twgl.createTexture(gl, {
    src: ctx.canvas,
    format: gl.LUMINANCE,  // because depth is only 1 channel
    min: gl.NEAREST,
    mag: gl.NEAREST,
  });

  function drawDepthImage(
      colorTex, depthTex, texWidth, texHeight,
      x, y, z) {
    var dstY = y + z;
    var dstX = x;
    var dstWidth = texWidth;
    var dstHeight = texHeight;

    var srcX = 0;
    var srcY = 0;
    var srcWidth = texWidth;
    var srcHeight = texHeight;

    gl.useProgram(programInfo.program);

    twgl.setBuffersAndAttributes(gl, programInfo, quadBufferInfo);

    // this matirx will convert from pixels to clip space
    var matrix = m4.ortho(0, gl.canvas.width, gl.canvas.height, 0, -1, 1);

    // this matrix will translate our quad to dstX, dstY
    matrix = m4.translate(matrix, [dstX, dstY, 0]);

    // this matrix will scale our 1 unit quad
    // from 1 unit to texWidth, texHeight units
    matrix = m4.scale(matrix, [dstWidth, dstHeight, 1]);

    // just like a 2d projection matrix except in texture space (0 to 1)
    // instead of clip space. This matrix puts us in pixel space.
    var texMatrix = m4.scaling([1 / texWidth, 1 / texHeight, 1]);

    // because were in pixel space
    // the scale and translation are now in pixels
    var texMatrix = m4.translate(texMatrix, [srcX, srcY, 0]);
    var texMatrix = m4.scale(texMatrix, [srcWidth, srcHeight, 1]);

    twgl.setUniforms(programInfo, {
      u_colorTexture: colorTex,
      u_depthTexture: depthTex,
      u_matrix: matrix,
      u_textureMatrix: texMatrix,
      u_depthOffset: 1 - (dstY - z) / 65536,
      u_depthScale: 1 / 256,
      u_dstSize: [dstWidth, dstHeight],
    });

    var numDstPixels = dstWidth * dstHeight;
    twgl.drawBufferInfo(gl, quadBufferInfo, gl.POINTS, numDstPixels);
  }

  // test render
  gl.enable(gl.DEPTH_TEST);

  var texWidth = 64;
  var texHeight = 64;

  // z is how much above/below ground
  function draw(x, y, z) {
    drawDepthImage(colorTexture, depthTexture, texWidth, texHeight , x, y, z);
  }

  draw(  0, 0, 0);  // draw on left

  draw(100, 0, 0);  // draw near center
  draw(113, 0, 0);  // draw overlapping

  draw(200, 0, 0);  // draw on right
  draw(200, 8, 0);  // draw on more forward

  draw(0, 60,  0);  // draw on left
  draw(0, 60, 10);  // draw on below

  draw(100, 60,  0);  // draw near center
  draw(100, 60, 20);  // draw below

  draw(200, 60, 20);  // draw on right
  draw(200, 60,  0);  // draw above
}

main();
<script src="https://twgljs.org/dist/2.x/twgl-full.min.js"></script>
<canvas></canvas>

Note that if it is too slow I don't actually think doing it in JavaScript in software is guaranteed to be too slow. You could use asm.js to make a renderer. You setup and manipulate the data for what goes where in JavaScript then call your asm.js routine to do software rendering.

As an example this demo is entirely software rendered in asm.js as is this one

If that ends up being too slow one other way would need some kind of 3D data for your 2D images. You could just use cubes if the 2D images are always cubic but I can already see from your sample picture those 2 cabinets require a 3D model because the top is few pixels wider than the body and on the back there's a support beam.

In any case, assuming you make 3D models for your objects you'd use the stencil buffer + the depth buffer.

  • For each object

    • turn on the STENCIL_TEST and DEPTH_TEST

      gl.enable(gl.STENCIL_TEST);
gl.enable(gl.DEPTH_TEST);

    • set the stencil func to ALWAYS, the reference to the iteration count, and the mask to 255

      var test = gl.ALWAYS;
var ref = ndx;  // 1 for object 1, 2 for object 2, etc.
var mask = 255;
gl.stencilFunc(test, ref, mask);

    • set the stencil operation to REPLACE if the depth test passes and KEEP otherwise

      var stencilTestFailOp = gl.KEEP;
var depthTestFailOp = gl.KEEP;
var bothPassOp = gl.REPLACE;
gl.stencilOp(stencilTestFailOp, depthTestFailOp, bothPassOp);

    • now draw your cube (or whatever 3d model represents your 2D image)

At this point the stencil buffer will have a 2D mask with ref everywhere the cube was drawn. So now draw your 2D image using the stencil to draw only where the cube was successfully drawn

  • Drawing the Image

    • Turn off the DEPTH_TEST

      gl.disable(gl.DEPTH_TEST);

    • Set the stencil function so we only draw where the stencil equals ref

      var test = gl.EQUAL;
var mask = 255;
gl.stencilFunc(test, ref, mask);

    • set the stencil operation to KEEP for all cases 

      var stencilTestFailOp = gl.KEEP;
var depthTestFailOp = gl.KEEP;
var bothPassOp = gl.KEEP;
gl.stencilOp(stencilTestFailOp, depthTestFailOp, bothPassOp);

    • draw the 2D image

      This will end up only drawing where the cube drew.

Repeat for each object.

You might want to clear the stencil buffer after every object or after every 254 objects and make sure ref is always between 1 and 255 because the stencil buffer is only 8 bits meaning that when you draw object 256 it will be using the same value as object #1 so if there are any of those values left in the stencil buffer there's a chance you might accidentally draw there.

 objects.forEach(object, ndx) {
    if (ndx % 255 === 0) {
      gl.clear(gl.STENCIL_BUFFER_BIT);
    }

    var ref = ndx % 255 + 1;  // 1 to 255

    ... do as above ...



回答2:


You can do this with separate pre-rendered Depth-Maps for every object. With additional Normal-Maps you can simulate Deferred Lighting, too. But this technique requires 3D creation for every object to render diffuse, normal and depthmaps for correct intersecting objects.

See the XNA-Demo at https://www.youtube.com/watch?v=-Q6ISVaM5Ww

Every object in this demo is just a rendered sprite with diffuse, normal and depthmap.

At the end of the video you see how it works. The author has an additional explanation & shader code-examples in his blog at https://infictitious.blogspot.de/2012/09/25d-xna-rpg-engine-some-technical.html

I think the same is possible with WebGL, too.



来源：https://stackoverflow.com/questions/41200514/webgl-3d-usage-for-depth-sorting-2d-objects

标签
html5
canvas
3d
WebGL
depth-buffer
易学教程内所有资源均来自网络或用户发布的内容，如有违反法律规定的内容欢迎反馈！
该文章没有解决你所遇到的问题?点击提问,说说你的问题,让更多的人一起探讨吧!