Three.js within web worker: Simulating animation without rendering to canvas - three.js
I have a hypothetical question:
Is it possible to simulate an animation of objects without rendering it to the canvas. I just want to capture objects' position using Vector.project(camera) and present it using CSS. And THREE.DeviceOrientationControls controls how the camera "view" the simulation.
I tried commenting THREE.WebGLRenderer, but it seems that THREE.PerpectiveCamera cannot update it's MatrixWorld property. Hence, the camera seems to not move and the Vector.project(camera) gives a static value. I do this because I need to put my three.js codes within a web worker.
Do I need still need to use THREE.WebGLRenderer to have a working simulation?
UPDATE:
I checked the following:
I digged deeper into ((three.scene.getObjectByName("one")).matrixWorld.getPosition()).project(three.camera);, I inspect the following values, having the above requirement (inside web worker, no renderer), using this example:
matrix: {"elements":{"0":3.2167603969573975,"1":0,"2":0,"3":0,"4":0,"5":2.1445069313049316,"6":0,"7":0,"8":0,"9":0,"10":-1.000100016593933,"11":-1,"12":5.4684929847717285,"13":2.1445069313049316,"14":-0.2000100016593933,"15":0}}
camera.projectionMatrix: {"elements":{"0":3.2167603969573975,"1":0,"2":0,"3":0,"4":0,"5":2.1445069313049316,"6":0,"7":0,"8":0,"9":0,"10":-1.000100016593933,"11":-1,"12":0,"13":0,"14":-0.2000100016593933,"15":0}}
camera.matrixWorld: {"elements":{"0":1,"1":0,"2":0,"3":0,"4":0,"5":1,"6":0,"7":0,"8":0,"9":0,"10":1,"11":0,"12":-1.7000000476837158,"13":-1,"14":0,"15":1}}
matrix.getInverse(camera.matrixWorld): {"elements":{"0":1,"1":0,"2":0,"3":0,"4":0,"5":1,"6":0,"7":0,"8":0,"9":0,"10":1,"11":0,"12":1.7000000476837158,"13":1,"14":0,"15":1}}
matrix.multiplyMatrices(camera.projectionMatrix, matrix.getInverse(camera.matrixWorld)): {"elements":{"0":3.2167603969573975,"1":0,"2":0,"3":0,"4":0,"5":2.1445069313049316,"6":0,"7":0,"8":0,"9":0,"10":-1.000100016593933,"11":-1,"12":5.4684929847717285,"13":2.1445069313049316,"14":-0.2000100016593933,"15":0}}
But, when normal (no modification), I inspect the following:
matrix: {"elements":{"0":3.2167603969573975,"1":0,"2":0,"3":0,"4":0,"5":2.1445069313049316,"6":0,"7":0,"8":0,"9":0,"10":-1.000100016593933,"11":-1,"12":5.4684929847717285,"13":2.1445069313049316,"14":-0.2000100016593933,"15":0}}
camera.projectionMatrix: {"elements":{"0":3.2167603969573975,"1":0,"2":0,"3":0,"4":0,"5":2.1445069313049316,"6":0,"7":0,"8":0,"9":0,"10":-1.000100016593933,"11":-1,"12":0,"13":0,"14":-0.2000100016593933,"15":0}}
camera.matrixWorld: {"elements":{"0":1,"1":0,"2":0,"3":0,"4":0,"5":-2.220446049250313e-16,"6":-1,"7":0,"8":0,"9":1,"10":-2.220446049250313e-16,"11":0,"12":-1.7000000476837158,"13":-1,"14":0,"15":1}}
matrix.getInverse(camera.matrixWorld): {"elements":{"0":1,"1":0,"2":0,"3":0,"4":0,"5":-2.220446049250313e-16,"6":1,"7":0,"8":0,"9":-1,"10":-2.220446049250313e-16,"11":0,"12":1.7000000476837158,"13":-2.220446049250313e-16,"14":1,"15":1}}
matrix.multiplyMatrices(camera.projectionMatrix, matrix.getInverse(camera.matrixWorld)): {"elements":{"0":3.2167603969573975,"1":0,"2":0,"3":0,"4":0,"5":-4.761761943205948e-16,"6":-1.000100016593933,"7":-1,"8":0,"9":-2.1445069313049316,"10":2.2206681307011713e-16,"11":2.220446049250313e-16,"12":5.4684929847717285,"13":-4.761761943205948e-16,"14":-1.2001099586486816,"15":-1}}
I noticed that the camera.matrixWorld property has significant difference in both condition. I do not understand what makes the difference.
Apparently, the following lines from THREE.WebGLRenderer.render are still needed to update camera.matrixWorld property:
scene.updateMatrixWorld();
camera.updateMatrixWorld();
camera.matrixWorldInverse.getInverse(vs._3.camera.matrixWorld);
Related
Three.js - is there a simple way to process a texture in a fragment shader and get it back in javascript code using GPUComputationRenderer?
I need to generate proceduraly in a shader a texture and get it back in my javascript code in order to apply it to an object. As this texture is not meant to change over time, I want to process it only once. I think that GPUComputationRenderer could do the trick but I don't figure out how and what is the minimal code that can achieve this.
I need to generate proceduraly in a shader a texture and get it back in my javascript code in order to apply it to an object. Sounds like you just want to perform basic RTT. In this case, I suggest you use THREE.WebGLRenderTarget. The idea is to setup a simple scene with a full-screen quad and a custom instance of THREE.ShaderMaterial containing your shader code that produces the texture. Instead of rendering to the screen (or default framebuffer), you render to the render target. In the next step, you can use this render target as a texture in your actual scene. Check out the following example that demonstrates this workflow. https://threejs.org/examples/webgl_rtt GPUComputationRenderer is actually intended for GPGPU which I don't think is necessary for your use case.
taking a 'snapshot' of babylonjs scene sometimes works, sometimes doesn't
I am trying to take a 'snapshot' of a babylon3d scene ... in other words: I am trying to clone a babylon3d canvas when a user presses a button, and then append the new <canvas> to the <body> .. Sometimes it works, but other times it does not. However, if I use a simple canvas (ie. by using fillRect), the cloning/appending always works as expected. I have set up a test on plunker to demonstrate my problem: plunker: press the button over and over again to see how sporadic it behaves when there is a babylon scene. AND NOTE: You can toggle between the simple canvas and the babylon canvas from within the _jquery(document).ready(...) handler. thanks, Shannon
This is because from version 2.3.0 of Babylonjs :
Engine now initialize WebGL with preserveDrawingBuffer = false by default.
You need to initialize the Engine by passing a {preserveDrawingBuffer: true} object as third parameter.
Forked plnkr
But this will unfortunately kill your canvas' performances.
See more about it here.
I'm not really a specialist of Babylonjs, and I didn't find a way to make a call from scene.render method where we could use the flag method proposed by
#CapsE. But there is a BABYLON.Tools.CreateScreenshot(engine, camera, size) method, which will make a downloadable png from your scene ; maybe this could help you.
Microsoft Edge WebGL lose context
I m creating a THREE.js (latest version, r71) app, and sometimes I need to manually delete the 3D scene to display classic 2D content. So, what I do is clearing all variables like scene or renderer, and killing the WebGL context using renderer.forceContextLoss() This method works fine on Firefox or Chrome, but it is not supported on Interner Explorer or Microsoft Edge, which cause multiple lags on my web page. I can't find a workaround to do this properly as I do in Firefox or Chrome. If someone has a tip, feel free to tell me :) Thanks
Try using two canvas tags and put one over the other as suggested here, put them in a parent div tag then use the following css. position:absolute; left:0px; top:0px; EDIT: In response to your comment, allow me to clarify my understanding. You want to draw 2d and 3d. You are resetting the webGL context. You can't have 2 contexts to the same canvas simultaneously and I assume you want to use a a different context for your 2d content. If this assumption is correct: I am suggesting that you have 2 canvases and contexts simultaneously, one with your webGL context and 3d content, and another with the context you want to use for 2d content. Then you overlay these canvases by using absolute position, such that it looks like there is one canvas. This means you never have to reload assets to switch contexts, just make sure you so the above canvas is transparent. Also if you wrap the canvas in a div tag, it will not be at the top left of the page. But If my assumption is wrong: perhaps you don't want to use another context and just wish to clear the screen, in which case you should not call forceContextLoss() If you wish to continue to use the webGL context, but clear the screen you should use clear()
In three.js r57, objects that are not double-sided do not show up
I am upgrading my application from three.js -r51 to -r57 (I got started before -r58 was released). When I did, I noticed that any of my 3D collada models that did not have in them a line like this: <extra><technique><double_sided>1</double_sided></technique></extra> did not render. It appears that the polygons are being culled. If I force _gl.disable( _gl.CULL_FACE ); my model shows up as expected. But why would it cull all of my faces all of the time? (even if I had the winding order backward, I should see the other side of the object, right?)
It turns out that my parameter to setFaceCulling wasn't right. I was passing false instead of a culling mode. With the r51 implementation, false happened to disable all culling. With the r57 implementation, it dumped me into the new default behavior ... which is to enable culling and set it to cull both front and back faces. So, I was able to fix it ... but I'm not sure I would have made that the default behavior. ;o)
camera.lookAt not called when THREE controls are being used
I am working on a program, that uses THREE.RollControls, when the user goes too far away from the center of the screen, they tend to get lost, so I am working on creating a function that reorients them, facing the center of the scene. What I had intened to do was simply call the following: camera.lookAt(scene.position) However, this has no affect. From what I was reading on different stack overflow questions specifically this: ThreeJS camera.lookAt() has no effect, is there something I'm doing wrong? It seems like their solution was to do the camera position change using the controls, rather then changing the camera itself. I do not believe there is any 'Target' in the Roll Controls, so I don't know how I can reset where the camera is looking at based on a THREE.Vector3() Is there a simple way to do this, or will I basically have to: So far I have 'attempted' to do the follow: - Calculate the difference of position of the camera with the position of the scene. - Normalize this vector - Subtract it from the direction forward of the camera - use this vector in controls.forward.add(thisVector) but this doesn't do at all what I want (probably because I have no idea what I'm doing) Thank you in advance for your time! Isaac
The same thing bugged me too about the RollControls but I took a different approach in solving the problem. Since the controls are in the example code (in r55) you can modify the controls, as they are not part of the core library. You can see my modifications at http://www.virtuality.gr/AGG/EaZD-WebGL/js/three.js/examples/js/controls/RollControls.js I introduced a local variable called mouseLook because I could not use the this.mouseLook. I initialized it to false and I only make it true when there is a button press i.e. when navigating in the scene. That solved my problem.