SuikaSibyl

SuikaSibyl / nqr

Public

Neural Quadrature Rule and Adaptive Sampling

19
0
100% credibility
Found May 15, 2026 at 19 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

This repository implements neural networks for improving numerical integration accuracy in graphics rendering tasks across various example scenarios like 1D functions, direct lighting, and distance fields.

How It Works

1
🔍 Discover cool math helpers

You stumble upon this project while looking for ways to make computer drawings more precise and faster.

2
📥 Grab the files

Download the main folder and extra picture data to try it out on your computer.

3
🎯 Pick a fun example

Choose a simple demo like basic math curves or light in a room to start playing with.

4
🧠 Train your smart assistant

Run a quick lesson so it learns to guess math answers better from sample points.

5
🚀 Test and compare

See side-by-side pictures showing how much sharper and quicker the results are.

✅ Enjoy perfect drawings

Your computer now solves tricky math for realistic images way faster than before.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 19 to 19 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is nqr?

nqr delivers neural quadrature rules and autoregressive adaptive sampling in Python with PyTorch and Slang shaders. It trains transformer models to reweight fixed Monte Carlo samples or generate smart new ones, slashing integration error for tasks like 1D benchmarks, transmittance, winding numbers, PDEs, UDF rendering, and direct illumination via SIByL. Users get low-variance estimates from few samples, with train/inference scripts and pre-trained checkpoints.

Why is it gaining traction?

Unlike plain MC, trapezoidal rules, or Gaussian processes, nqr's adaptive neural approach yields 10-100x variance cuts in demos, no heuristics needed. GitHub neural operator fans like its autoregressive sampling akin to neural forecast or neural prophet for integration. Tied to a SIGGRAPH 2026 paper, it hooks graphics devs with runnable examples beating baselines.

Who should use this?

Rendering engineers tuning Monte Carlo for path tracers or volumes. Neural operator researchers tackling adaptive quadrature in PDEs or UDFs. Python graphics hackers needing efficient integrators without nqrman-level complexity.

Verdict

Worth forking for neural rendering experiments—solid examples, CUDA-ready. 19 stars and 1.0% credibility signal early research code; docs are README-focused, test light, but baselines prove the gains.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.