Skip to content

Latest commit

 

History

History
329 lines (209 loc) · 18 KB

README.md

File metadata and controls

329 lines (209 loc) · 18 KB

Three.js Journey

This is a repository hosting code used throughout my time learning Three.js from Bruno Simon's Three.js course. Below are some notes about the basic usages of Three.js.

Basic parts to a scene

  • Scene which is essentially the set and made by const scene = new THREE.Scene();
  • Objects that make up the scene; these consist of a geometry (shape), and material
  • Camera which views the scene and is made up of a perspective, and position
  • Canvas which was added to the HTML and is rendered in the JS through a renderer; the scene and camera are added to the renderer

Objects (Mesh)

  • made up of a geometry and material

Example:

const cube1 = new THREE.Mesh(
  new THREE.BoxGeometry(1, 1, 1),
  new THREE.MeshBasicMaterial({ color: 0xff0000 })
);

Some methods to manipulate meshes:

  • mesh.position.x/y/z translates the mesh to different points in space
  • mesh.scale.x/y/z scales the mesh to be bigger or smaller along an axis
  • mesh.rotation.x/y/z rotates the mesh along an axis in radians

Objects can be added to a new THREE.Group() so that the entire group is manipulated rather than applying the same transformations to each object individually.

Basic Animation

A simple way to make a Three.js animation is to make an infinite loop function that manipulates the object or camera and then rerenders the scene using window.requestAnimationFrame().

The problem is that different computers have different framerates which makes the speed at which something rerenders inconsistent among computers.

Possible solutions:

  • use the Date function in JS to regulate the time at which it rerenders
  • use the THREE.Clock() function

There is also another library called GSAP which can be used to animate objects since it works with Three.js.

Cameras

There are different types of cameras:

  1. ArrayCamera - used to render things like split screens
  2. StereoCamera - used to render things like VR
  3. CubeCamera - used to render environment maps for reflection or shadow map
  4. OrthographicCamera - renders scene without perspective
  5. PerspectiveCamera - similar to real-life camera with perspective

The PerspectiveCamera has 4 different parameters:

  1. Field of view - the vertical amplitude angle in degrees; small angle creates long scope effect while wide angle is similar to fish eye effect; typical values are 45-75
  2. Aspect ratio - can typically use the canvas width and height which should be saved in an object since you use it multiple times
  3. Near (3) and far (4) - how close and how far the camera can see (range); try not to use too close and too far otherwise you'll get z-fighting where objects render into each other; typically use 0.1 and 100

The OrthographicCamera is different from the PerspectiveCamera and has 6 different parameters: left, right, top, bottom, near, and far. The first four indicate how far the camera can see in those directions.

Controls

There are some pre-made controls:

  • DeviceOrientationControls - used to retrieve device orientation if rotation is allowed
  • FlyControls - lets you rotate all 3 axes, go forward, and go backwards
  • FirstPersonControls - fixed up axis like a flying bird view where the bird can't do a barrel roll
  • PointerLockControls - hides the cursor and keeps it centered, allowing you make FPS games
  • OrbitControls - left click = rotate, right click = panning, mosue wheel = zooming
  • TrackballControls - similar to OrbitControls but has no limit
  • TransformControls - doesn't touch the camera, attaches to object and helps to move the object
  • DragControls - doesn't touch the camera, moves objects on a plane facing the camera by dragging and dropping them

OrbitControls

  • can use target to change the postion of where the camera is targeting for x, y, z ex. controls.target.y = 2
  • damping will smooth the animation by adding acceleration and friction which can be enabled by controls.enableDamping = true

Geometries

There is the BoxGeometry which has been used up until now, but what is a geometry? Geometries are made up of vertices and faces. These are used to make meshes and can be used to make particles.

There are a lot of built in geometries in Three.js which start with the following and end in Geometry:

  • Box - box
  • Plane - rectangle plane
  • Circle - disk or part of a disk
  • Cone - cone or part of a cone
  • Ring - flat ring or portion of flat circle
  • Torus - donut type ring
  • TorusKnow - knot thing
  • Dodecahedron - 12 faced sphere
  • Octahedron - 8 faced sphere
  • Tetrahedron - 4 faced sphere
  • Icosahedron - sphere made of triangles with the same size
  • Sphere - basically a sphere with the quads
  • Shape - shape based on a path
  • Tube - tube following a path
  • Extrude - making a extrusion based on a path
  • Lathe - creates a vase
  • Text - creates 3D text

As for the parameters of these built in geometries, there is the width, height, depth, widthSegments, heightSegments, and depthSegments. It works on spheres to change the segment parameters so that it is more smooth. By default it is 1, so if you set it to more, then there will be more segments and appear more smooth.

We can use BufferGeometry to make our own shapes. This involves creating an array with the locations of the points, and the number of points used per shape in the BufferAttribute since you can make multiple of the same shape using the array and methods.

Textures

  • images that cover the surface of geometries

There are different types of textures:

  • Color which takes the pixels of the texture and applies it to the geometry
  • Alpha which takes a grayscale image where while is visible and black is not
  • Height which is a grayscale image that moves the vertices to create some relief
  • Normal which will add small details but not move the vertices
  • Ambient occlusion which is a grayscale image that will fake shadow on surface crevices
  • Metalness which is a grayscale image that will show metallic parts as white and non-metallic as black helping to create a reflection
  • Roughness is a grayscale image that goes with metalness and helps to dissipate the light

PBR (physically based rendering) principles is one of the regrouping techniques that follow real life situations to get more realistic renderings and results.

  • use the .TextureLoader() to load the image, it also has progress function for when the loading is happening and an error funnction if something goes wrong

  • use .LoadingManager() to help you manage loading a number of textures into your project

  • use UV Unwrapping when we need to wrap a texture around a weird shaped object; if we have to create our own geometry, we have to also do the UV unwrapping ourselves through the 3D software

Mipmapping is a technique that consists of creating half a smaller version of a texture again and again until you get a 1x1 texture. There are two filter algorithms: minification and magnification filters. Only use the mipmaps for the minFiler property and you don't need it if you are using THREE.NearestFilter, so you can deactivate it using colorTexture.generateMipmaps = false.

Minification filter is when the pixels of texture are smaller than the pixels of the render, so texture is too big for the surface and has a couple of different filters:

  • THREE.NearestFilter
  • THREE.LinearFilter
  • THREE.NearestMipmapNearestFilter
  • THREE.NearestMipmapLinearFilter
  • THREE.LinearMipmapNearestFilter
  • THREE.LinearMipmapLinearFilter

Using the THREE.NearestFilter results in some weird effects happening called moiré patterns which render crazy things.

Magnification filter is similar to minification filter but it makes it beeger so that the texture is too small for the surface to cover, and thus is more blurry. This one has two filters:

  • THREE.NearestFilter - kinda like the Minecraft pixelated effect, gets the best performance out of all the options
  • THREE.LinearFilter

With textures, need to keep in mind the weight, size, and data. For weight, the images need to be downloaded upon visiting the site, so need to make it as compressed as possible. For size, reduce the pixel size as much as possible and needs to be a power of 2. For data, textures support transparency, and need to use pngs for lossless compression.

Some places to find textures include:

Materials

  • used to put color on each visible picture of a geometry

MeshBasicMaterial is the most basic material and using the map property will let you add a texture while the color property lets you add a color class

  • one can combine the color and map properties to tint the texture with a colour
  • the wireframe property lets you see it as a geometry
  • one can also change the opacity of a material, first set transparent = true and then set opacity = #
  • alphaMap is a grayscale texture that controls the opacity across the surface (black: fully transparent; white: fully opaque)
  • side lets you choose which side is visible (either front, back or double sides)

MeshNormalMaterial lets you see the rainbowy colours in the normal relative's orientation to the camera; with this you can also use flatShading which flattens the faces

MeshMatcapMaterial is an illuminated looking material that is performative MeshDepthMaterial will colour the geometry in white if it is close to the camera and black if it is far

There are some that need to use a light source in order to render properly:

  • MeshLambertMaterial is a material that renders based on lighting in the scene
  • MeshPhongMaterial uses the same things as the previous but with less strange patterns and can change shininess and specular properties
  • MeshToonMaterial makes it look like it came out of a cartoon and you can use the gradientMap property to declare colours for the shadow and for the light in which we need to ensure we use a large enough gradient texture

MeshStandardMaterial uses the standard physically based rendering principles and you can change the metalness and roughness properties

  • can use aoMap property to add shadows where the textures are dark by giving the UV coordinates
  • the displacementMap properties will move the vertices to create true relief and need to make sure we have enough displacement and more subdivisions by changing the scale
  • normalMap will fake the normal orientation and add details on the surface regardless of the subdivision; there is also the metalnessMap and roughnessMap

An environment map is an image of what's happening in the surrounding scene and is used to add reflection or refraction to the objects, makes it similar to a mirror or shiny object. A good resource for environment maps are HDRIHaven and you need to convert it to a cube map using https://matheowis.github.io/HDRI-to-CubeMap/.

Fonts

In three.js we can make 3D text. First, we need to make a .json of the font face and then add it to the project. Then, we need to load it in. Afterwards, we can use the same functions on objects to change the materials.

We can use textGeometry.center() to centre the text.

A good resource for getting a matcap is https://github.com/nidorx/matcaps.

Lighting

AmbientLight applies omnidirectional lighting on all the geometries of the scene, taking parameters such as color and intensity.

DirectionalLight has a sun effect with sun rays going in parallel.

HemisphereLight is similar to AmbientLight but has a different colour coming from the sky and from the ground. This makes it so that there are two colour parameters and an intensity one.

PointLight is like a lighter where the light is very small and is uniform. You can change the fade distance and how quickly it fades using the distance and decay properties.

RectAreaLight is similar to large rectangle lights and is similar to directional lighting and diffuse light. There are the color, intensity parameters, but also width of the rectangle and height.

SpotLight is like a flashlight and has the first two parameters and distance, angle, penumbra, and decay parameters.

Some performace considerations: Minimal cost - AmbientLight, HemisphereLight

Moderate cost - DirectionalLight, PointLight

High cost - SpotLight, RectAreaLight

Baking is when you put the light into the texture, but you can't move the lights.

Shadows

With lights we also need shadows. The back part of objects that are dark are called core shadows. The drop shadows are the shadows created on other objects.

When doing a render, Three.js will do a render for all lights that cast shadows. These will simulate the lights seens as if it were a camera. When the light renders, MeshDepthMaterial replaces all mesh materials. It's then stored as textures and named shadow maps.

To set it up, you have to enable the shadowMap, set up castShadow and receiveShadow on the objects that have shadows, and then make the light cast the shadow with castShadow.

The problem with baking the shadow into a texture is that it is not dynamic.

Particles

Parts to make a particle: BufferGeometry, PointsMaterial and Points

PointsMaterial is a special type of material to create particles. It lets us use size, and sizeAttenuation to specify if distant particles should be smaller than close particles.

alphaTest is a value between 0 and 1 that lets WebGL know when and when not to render pixel based on transparency.

depthTest lets you test to see if what's being drawn is closer than what's already drawn. Doesn't work well when there are other objects in the scene though.

A depth buffer is the depth of what's being drawn is stored.

depthWrite lets us tell WebGL to not write particles in that depth buffer. And works ok in our case with particles and other objects in the scene.

blending is a property that lets us not only to draw to a pixel, but to add the color of that pixel to the color of the pixel already drawn. So in our case, it lets us see the halo effect on top of another halo, making it super saturated in the places that they overlap. This effect reduces the performance though, so make sure to use less particles.

Raycaster

Shoots a ray in a direction to test to see which objects intersect with the ray. The two main methods are intersectObject and intersectObjects.

Each item of that returned array contains much useful information:

  • distance: the distance between the origin of the ray and the collision point.
  • face: what face of the geometry was hit by the ray.
  • faceIndex: the index of that face.
  • object: what object is concerned by the collision.
  • point: a Vector3 of the exact position in 3D space of the collision.
  • uv: the UV coordinates in that geometry.

Mouse events like mouseEnter and mouseLeave are not supported so we have to reproduce those methods if we want to use them.

Physics

Some 3D physics libraries:

  • Ammo.js
  • Cannon.js
  • Oimo.js

Some 2D physics libraries:

  • Matter.js
  • P2.js
  • Planck.js
  • Box2D.js

To use Cannon.js, we make a world const world = new CANNON.World(); and then we can set things like gravity. To update the World with the physics we have to use the step(...) function which can be explained further here.

Create shapes and bodies in Cannon.js. Then you can add ContactMaterials to change the material properties and help things bounce or have friction.

There's some options to apply forces:

  • applyForce
  • applyImpulse
  • applyLocalForce
  • applyLocalImpulse

To test collisions between objects, we can use a broadphase which is a rough sorting of the bodies before testing. This is check if two bodies are close enough to each other to warrant testing. There are 3 algorithms including:

  • NaiveBroadPhase (default) - every body against every other body
  • GridBroadphase - only tests bodies against other bodies in the same grid space
  • SAPBroadphase (recommended) - tests bodies on arbitrary axes during multiple stages

Since all bodies get tested even if they're not moving, we can use a feature called sleep which means an object will not get tested if it is moving really slowly or not moving at all. Thus, the object won't get tested until it is hit with a lot of force through code or by another object. You just need to turn it on.

We can listen to events that occur on a body. Some common events are collide, sleep, and wakeup.

You can add constraints like HingeConstraint, DistanceConstraint, LockConstraint, and PointToPointConstraint.

Works help to put a part of the code on a different thread so that you can spread the load to your CPU and not cause a single thread to be overloaded. This is a good example here.

The real Cannon.js is not updated so we use cannon-es which is maintained here:

Ammo.js is another good choice if more features are needed. It's more popular and has more examples of Three.js. There are also more performance enhancing features and WebAssembly.

Physijs is also another good choice since it has workers natively. It gets complicated quickly when adding more features though.

Importing Models

Common model formats:

  • OBJ
  • FBX
  • STL
  • PLY
  • COLLADA
  • 3DS
  • GLTF (very common)

When looking at the GLTF format, it's good to use the glTF-default so that we can alter textures and lighting. If we only want one file per model and don't care about customization, glTF-Binary is good.

Draco compression sounds like a good idea since the geometries are lighter, but we have to load numerous things in, and it takes time to decode a compressed file and can result in a short freeze at the start of the program.

You can also load animations that come with the models using the animation mixer.