Giter Site home page Giter Site logo

gfxexp's Introduction

Graphics Experiments

適当にグラフィックス関連の論文などを読んで実装・検証したものを置きます。

このリポジトリを正しくcloneするにはGit LFSのインストールが必要です。

I'll randomly put something for implementing/validating graphics papers here.

You need to install Git LFS to correctly clone this repository.

実装 / Implementations

ReSTIR

Spatiotemporal reservoir resampling for real-time ray tracing with dynamic direct lighting
https://research.nvidia.com/publication/2020-07_Spatiotemporal-reservoir-resampling

ReSTIRでは、Resampled Importance Sampling (RIS), Weighted Reservoir Sampling (WRS)、そして複数のReservoirを結合する際の特性を利用することで、プライマリーヒットにおいて大量の発光プリミティブからの効率的なサンプリングが可能となります。

ReSTIR enables efficient sampling from a massive amount of emitter primitives at primary hit by Resampled Importance Sampling (RIS), Weighted Reservoir Sampling (WRS) and utilizing the property of combining multiple reservoirs.

example Amazon Lumberyard Bistro (Exterior) from Morgan McGuire's Computer Graphics Archive

ReGIR

Chapter 23. "Rendering Many Lights with Grid-based Reservoirs", Ray Tracing Gems II
https://www.realtimerendering.com/raytracinggems/rtg2/index.html

ReGIRでは、ReSTIRと同様にStreaming RISを用いて大量の発光プリミティブからの効率的なサンプリングが可能となります。ReSTIRとは異なり、セカンダリー以降の光源サンプリングにも対応するため、Reservoirをワールド空間のグリッドに記録し、2段階のStreaming RISを行います。

ReGIR enables efficient sampling from a massive amount of emitter primitives by using streaming RIS similar to ReSTIR. Unlike ReSTIR, ReGIR stores reservoirs in a world space grid and performs two-stage streaming RIS to support light sampling after secondary visibility.

  • Basic Implementation (Uniform Grid, Temporal Reuse)
  • Advanced Items
    • Diffuse + Glossy BRDF
    • Environmental Light
    • Scrolling Clipmap or Sparse Grid using Hash Map
    • ReGIR + Multiple Importance Sampling (Impossible?)

example Amazon Lumberyard Bistro (Interior) from Morgan McGuire's Computer Graphics Archive

Neural Radiance Caching

Real-time Neural Radiance Caching for Path Tracing
https://research.nvidia.com/publication/2021-06_Real-time-Neural-Radiance

Path Tracing + Neural Radiance Cache (NRC)は、ある経路長より先から得られる寄与をニューラルネットワークによるキャッシュからの値によって置き換えることで、少しのバイアスと引き換えに低い分散の推定値(、さらにシーンによっては短いレンダリング時間)を実現します。NRCは比較的小さなネットワークであり、トレーニングはレンダリングの最中に行うオンラインラーニングとすることで、「適応による汎化」を実現、推論の実行時間もリアルタイムレンダリングに適した短いものとなります。

Path Tracing + Neural Radiance Cache (NRC) replaces contributions given from beyond a certain path length by a value from the cache based on a neural network. This achieves low variance estimates at the cost of a little bias (, and additionally rendering time can even be reduced depending on the scene). NRC is a relatively small network, and training is online learning during rendering. This achieves "generalization via adaptation", and short inference time appropriate to real-time rendering.

  • Basic Implementation (based on simple path tracing, frequency/one-blob input encoding)
  • Advanced Items
    • Combine with many-light sampling techniques like ReSTIR/ReGIR
    • Add multi-resolution hash grid input encoding:
      "Instant Neural Graphics Primitives with a Multiresolution Hash Encoding"
      https://nvlabs.github.io/instant-ngp/

example Zero-Day from Open Research Content Archive (ORCA)

SVGF

Spatiotemporal Variance-Guided Filtering: Real-Time Reconstruction for Path-Traced Global Illumination
https://research.nvidia.com/publication/2017-07_spatiotemporal-variance-guided-filtering-real-time-reconstruction-path-traced

SVGFはパストレーシングなどによって得られたライティング結果を、物体表面のパラメターを参照しつつ画像空間でフィルタリングします。各ピクセルのライティングの分散を時間的・空間的にトラッキングし、分散が小さな箇所では小さなフィルター半径、分散が大きな箇所では大きなフィルター半径とすることで画像のぼけを抑えつつレンダリング画像における視覚的なノイズを低減します。フィルターにはà-trous Filterを用いることで大きなフィルタリング半径を比較的小さなコストで実現します。

SVGF filters the lighting result in screen-space obtained by methods like path tracing with references to surface parameters. It tracks the variance of the lighting for each pixel in spatially and temporally, then uses smaller filter radii at lower variance parts and larger filter radii at higher variance parts to get rid of perceptual noises from the rendered image while avoiding excessive blurs in the image. It uses an à-trous filter so that large filter radii can be used with relatively low costs.

  • Basic Implementation (temporal accumulation, SVGF, temporal AA)
  • Advanced Items

example Crytek Sponza from Morgan McGuire's Computer Graphics Archive

その他 / Miscellaneous

OptiX/CUDAのラッパーとしてOptiX Utilityを使用しています。

Programs here use OptiX Utility as OptiX/CUDA wrapper.

動作環境 / Confirmed Environment

現状以下の環境で動作を確認しています。
I've confirmed that the program runs correctly in the following environment.

  • Windows 10 (21H2) & Visual Studio Community 2022 (17.3.4)
  • Core i9-9900K, 32GB, RTX 3080 10GB
  • NVIDIA Driver 516.94 (Note that versions around 510-512 had several OptiX issues.)

動作させるにあたっては以下のライブラリが必要です。
It requires the following libraries.

  • CUDA 11.6 Update 2
  • OptiX 7.5.0 (requires Maxwell or later generation NVIDIA GPU)

オープンソースソフトウェア / Open Source Software


2022 @Shocker_0x15

gfxexp's People

Contributors

shocker-0x15 avatar

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.