MatForge - Advanced Material Rendering System

MatForge Banner

Overview

MatForge is an advanced material rendering system that implements FOUR complementary SIGGRAPH papers (2023-2024) in a unified Vulkan ray tracing pipeline. Our goal is to create a production-quality material authoring and rendering workflow that combines state-of-the-art techniques for sampling, geometry, and shading.

Project Context

Course: CIS 5650 GPU Programming and Architecture
Semester: Fall 2025
Institution: University of Pennsylvania
Team: 3 Students (Yiding Liu, Cecilia Chen, Xiaonan Pan)
Timeline: November 3 - December 7, 2025 (5 weeks)
Platform: Vulkan 1.3 with ray tracing extensions
Base Framework: Forked from NVIDIA nvpro-samples/vk_gltf_renderer

Features

Rendering Pipeline Architecture

┌─────────────────────────────────────────────────────┐
│              MatForge Rendering Pipeline            │
├─────────────────────────────────────────────────────┤
│                                                     │
│  1. QOLDS - Sampling Foundation                     │
│     └─ Generate low-discrepancy random numbers      │
│        ↓                                            │
│  2. RMIP - Geometry Detail                          │
│     └─ Ray-trace displacement maps directly         │
│        ↓                                            │
│  3. Bounded VNDF - Direction Sampling               │
│     └─ Efficient importance sampling                │
│        ↓                                            │
│  4. Fast-MSX - BRDF Evaluation                      │
│     └─ Multiple scattering approximation            │
│        ↓                                            │
│  5. MONTE CARLO INTEGRATION                         │
│     └─ Combine: f(ωᵢ, ωₒ) × L(ωₒ) × cos(θ) / PDF    │
│                                                     │
└─────────────────────────────────────────────────────┘

Implemented Techniques

✅ Quad-Optimized Low-Discrepancy Sequences (QOLDS)

Paper: Ostromoukhov et al., ACM SIGGRAPH 2024 Status: Complete

Base-3 Sobol’ sequences with (1,4)-sequence property
Owen scrambling for randomization
15-30% variance reduction vs. standard sampling
Negligible performance overhead
47 dimensions × 243 max points (3^5)

Benefits: Foundation layer providing better random samples for all Monte Carlo integration tasks.

✅ Fast Multiple Scattering Approximation (Fast-MSX)

Paper: ACM SIGGRAPH 2023 Status: Complete

Relaxed V-cavity model for GGX materials
Modified distribution and geometry terms
Additive multi-scatter BRDF term
100× better energy conservation at high roughness
~5% performance overhead

Benefits: Fixes dark edges on rough materials, improves physical accuracy of GGX BRDF.

🚧 RMIP (Rectangular MinMax Image Pyramid)

Paper: Thonat et al., ACM SIGGRAPH Asia 2023 Status: In Progress (GPU data structure complete, ray tracing integration underway)

Tessellation-free displacement ray tracing
Hierarchical min-max pyramid for oblong bounding
Inverse mapping: ray ↔ texture space
11× faster than TFDM, 3× less memory than tessellation

Benefits: Add high-frequency geometric detail without mesh complexity.

📋 Bounded VNDF Sampling

Paper: Eto & Tokuyoshi (AMD), ACM SIGGRAPH Asia 2023 Status: Planned for Milestone 2

Tighter spherical cap bound for GGX sampling
15-40% variance reduction for rough surfaces (α = 0.6-1.0)
Minimal overhead (<1%)

Benefits: Fewer rejected samples for rough materials, faster convergence.

Gallery

Milestone 1 Progress

QOLDS Implementation QOLDS sampling integration with GUI toggle and console logging

RMIP Structure RMIP hierarchical min-max pyramid structure

Fast-MSX Showcase Fast-MSX multi-scattering approximation improving energy conservation

Team

Team Member	Role	Technique	Status
Yiding Tian	Sampling Specialist	Quad-Optimized LDS	✅ Complete
Cecilia Chen	Geometry Specialist	RMIP Displacement	🚧 In Progress
Xiaonan Pan	Material Specialist	Fast-MSX + Bounded VNDF	✅ Fast-MSX Complete

Milestones

Milestone 1: November 12, 2025 ✅

Goal: Individual techniques working (foundations)

Achievements:

✅ QOLDS fully integrated (700 LOC)
✅ RMIP GPU data structure builder (800 LOC)
✅ Fast-MSX implementation (350 LOC)
✅ Total: ~1,850 lines of production code

View Milestone 1 Report →

Milestone 2: November 24, 2025 🎯

Goal: Full pipeline integration + material system

Target Deliverables:

Complete RMIP ray tracing integration
Bounded VNDF implementation
Material library (7+ materials)
Performance benchmarks
Comparison modes (toggle each technique)

Final Presentation: December 7, 2025 📅

Goal: Production features + comprehensive analysis

Final Deliverables:

Complete 4-technique pipeline
Material parameter editor
Demo video
Technical documentation
Performance analysis

Documentation

Quick Links

GitHub Repository
README.md - Quick start and build instructions
Milestone 1 Report

Papers

Quad-Optimized LDS: Ostromoukhov et al., “Quad-Optimized Low-Discrepancy Sequences”, ACM SIGGRAPH 2024
RMIP: Thonat et al., “Displacement ray-tracing via inversion and oblong bounding”, ACM SIGGRAPH Asia 2023
Bounded VNDF: Eto & Tokuyoshi, “Bounded VNDF Sampling for Smith-GGX Reflections”, ACM SIGGRAPH Asia 2023
Fast-MSX: “Fast Multiple Scattering Approximation”, ACM SIGGRAPH 2023

Performance Targets

Resolution	Quality	Target FPS	Hardware
1080p	Interactive	60	RTX 4070
1080p	Balanced	30	RTX 4070
1440p	Quality	20	RTX 4070

Expected Improvements:

QOLDS: 15-30% variance reduction
RMIP: 11× faster than tessellation
Bounded VNDF: 15-40% fewer rejected samples
Fast-MSX: 100× energy conservation improvement

Build Instructions

Quick Start

# Clone repository
git clone https://github.com/matforge/MatForge.git
cd MatForge

# Configure and build (Windows)
cmake -B build -S . -DUSE_DLSS=ON
cmake --build build --config Release

# Run
.\_bin\Release\vk_gltf_renderer.exe

Requirements:

Windows 10/11
NVIDIA RTX GPU (RTX 20-series or newer)
Vulkan SDK 1.3+
CMake 3.25+
C++20 compiler (MSVC 2022 / GCC 11+)

For detailed build instructions, see README.md.

Acknowledgments

Instructor: Shehzan Mohammed (University of Pennsylvania)
Base Framework: NVIDIA nvpro-samples
Papers: SIGGRAPH 2023-2024 authors
Course: CIS 5650 GPU Programming

Contact

GitHub: matforge/MatForge
Course Forum: Ed Discussion
LinkedIn: CIS 5650 Alumni Group

Last Updated: November 18, 2025