Advanced Global Illumination in Unity 5 - Unite Europe 2015

Advanced Global Illumination in Unity 5 - Unite Europe 2015

(KASPER STOMRM)

BakedGI

Lightmaps and light probes

Precomputedrealtime GI

 Lightmaps and light probes

Bakedand realtime reflection probes

 

Those3 components can freely be combined.

 

Whatdoes precomputed realtime GI mean?

 

Freelychange all light properties in the player

 ADD/remove lights

 Change light properties

Freelymodify material reflectance and emission properties of static geometry (Metashader pass)

 

Freelychange the environment lighting

 

Lightingworkflow

 

Hashingand caching

Filebased

Hashinputs and compute missing outputs

  Robust

  General

  Easyto distribute

  Cacheable

  Async

Allowsfor “Auto mode”  and  “snapshot mode”

LightmappingModes

Performance

  Shading performance

  Memory footprint

 

Non-Directional

  Flat disuse

  Tex0: diffuse response

  Fast

  Lowest memory footprint

  Good for mobile

 

Directional

 Normalmapped diffuse

Tex0:diffuseresponse

Tex1:directionality(fixed 4)

  Dominant light direction

  Length= directionality

  Rebalancing coefficient

Combineddirectionality for direct and in direct

Relativelyfast, twice the memory

Directionalspecular

Diffuseand specular

  Both normalmapped

Tex0:intensity

Tex1:directionality(fixed3)

  Dominant light direction

  Length = directionality

 

Metapass

Customshader pass using the GPU

Decouplethe albedo and emissive Enlighten uses, from the regular shader passes

Modifyalbedo (material reflectance)

Changeemission of static geometry

Additiveloading of lighting data

Bakinglighting data can be done in two ways

  Separately (separate levels , no sharedlighting)

  Together (multiple levels can affect eachother

 Preview available in the Unity 5.2 alpha

 

Enlightenupgrade from 2.22 to 3.02

Performanceimprovements

New feature:real-time transparency

  Remove objects at runtime

  Destruction , frosted windows, doors etc.

Improvedcubemaps

Probelit objects currently not visible to Enlighten

  Enlighten 3.0 includes all immovable objectsin clustering

  Which means they can act as emitters!

 

Bestpractices

Set arealtime resolution that makes sense for you scene

Use thefast emissive API (DynamicGI .SetEmissive)

Add acustom meta pass to change bounce lighting and emission without affection theregular shader passes.

Add glowby attaching lights to dynamic objects

Use emissivesurfaces as area lights.

 

AdamSimonar

NVYVE工作室发布了第一人称恐怖游戏《帕梅拉(P.A.M.E.L.A.)》

Geometryimport and setup

 

As you’remodelling/designing ,try to plan what needs to be pre-computed vs using lightprobes.

Generally,only large architectural elements should be set as static (use namingconventions such as _LMS to denote pre_computed geometry.

 

Keepin mind each object’s desired effect on the scene; in the below example, thegreen tubes are physically very small ,but have a large impact on the scene’slighting so they were set as static.

 

Performanceand material tips

 

 For object that need to have emission updatedat runtime, make sure to keep them as separate meshes that can use a flat colorvs. complex texture map for emission updates

 

 Use dynamicGI.SetEmissive rather thanDynamicGI.UpdateMaterials whenever possible for better performance

 

Updatinglighting and materials is the expensive part and can incur big GPU spikes onlarge scenes

  Don’t set unnecessary lights as dynamic, e.g.enemy player flashlights

 

Stayroughly within physically based values when creating materials

   Overly bright or dark materials can produceincorrect ,odd looking GI

 

 

深度学习是机器学习的一个子领域,它基于人工神经网络的研究,特别是利用多层次的神经网络来进行学习和模式识别。深度学习模型能够学习数据的高层次特征,这些特征对于图像和语音识别、自然语言处理、医学图像分析等应用至关重要。以下是深度学习的一些关键概念和组成部分: 1. **神经网络(Neural Networks)**:深度学习的基础是人工神经网络,它是由多个层组成的网络结构,包括输入层、隐藏层和输出层。每个层由多个神经元组成,神经元之间通过权重连接。 2. **前馈神经网络(Feedforward Neural Networks)**:这是最常见的神经网络类型,信息从输入层流向隐藏层,最终到达输出层。 3. **卷积神经网络(Convolutional Neural Networks, CNNs)**:这种网络特别适合处理具有网格结构的数据,如图像。它们使用卷积层来提取图像的特征。 4. **循环神经网络(Recurrent Neural Networks, RNNs)**:这种网络能够处理序列数据,如时间序列或自然语言,因为它们具有记忆功能,能够捕捉数据中的时间依赖性。 5. **长短期记忆网络(Long Short-Term Memory, LSTM)**:LSTM 是一种特殊的 RNN,它能够学习长期依赖关系,非常适合复杂的序列预测任务。 6. **生成对抗网络(Generative Adversarial Networks, GANs)**:由两个网络组成,一个生成器和一个判别器,它们相互竞争,生成器生成数据,判别器评估数据的真实性。 7. **深度学习框架**:如 TensorFlow、Keras、PyTorch 等,这些框架提供了构建、训练和部署深度学习模型的工具和库。 8. **激活函数(Activation Functions)**:如 ReLU、Sigmoid、Tanh 等,它们在神经网络中用于添加非线性,使得网络能够学习复杂的函数。 9. **损失函数(Loss Functions)**:用于评估模型的预测与真实值之间的差异,常见的损失函数包括均方误差(MSE)、交叉熵(Cross-Entropy)等。 10. **优化算法(Optimization Algorithms)**:如梯度下降(Gradient Descent)、随机梯度下降(SGD)、Adam 等,用于更新网络权重,以最小化损失函数。 11. **正则化(Regularization)**:技术如 Dropout、L1/L2 正则化等,用于防止模型过拟合。 12. **迁移学习(Transfer Learning)**:利用在一个任务上训练好的模型来提高另一个相关任务的性能。 深度学习在许多领域都取得了显著的成就,但它也面临着一些挑战,如对大量数据的依赖、模型的解释性差、计算资源消耗大等。研究人员正在不断探索新的方法来解决这些问题。
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值