High-Quality Real-Time Rendering Using Subpixel Sampling Reconstruction

[AAAI 2024] High-Quality Real-Time Rendering Using Subpixel Sampling Reconstruction

Abstract

Generating high-quality, realistic rendering images for real-time applications generally requires tracing a few samples-per-pixel (spp) and using deep learning-based approaches to denoise the resulting low-spp images. Existing denoising methods have yet to achieve real-time performance at high resolutions due to the physically-based sampling and network inference time costs. In this paper, we propose a novel Monte Carlo sampling strategy to accelerate the sampling process and a corresponding denoiser, subpixel sampling reconstruction (SSR), to obtain high-quality images. Extensive experiments demonstrate that our method significantly outperforms previous approaches in denoising quality and reduces overall time costs, enabling real-time rendering capabilities at 2K resolution.

Preparation

This repo is tested with Ubuntu 20.04, python==3.7/3.8, pytorch==1.4.0 and cuda==10.1.

Please download SSR dataset and organize the data as follows, then set path in the settings.py with the corresponding data location.

Subpixel dataset
├── spp32768_train
|  └── [scene name]
|  └── ...
├── spp32768_test
|  └── [scene name]
|  └── ...
├── spp32768_val
|  └── [scene name]
|  └── ...
...

Here we provide a detailed introduction to the G-buffer features.

	R	G	B	A
Color	albedo
Normal	normal			AlphaMode
Position	position			HitModelFlag
Emissive	emissive			AO
PBR	bDoubleSided	roughness	metallic	AlphaCutoff
FWidth	N Width	depth	position	PrimitiveID
	R16	G16
Velocity	x	y	ViewDist	Mesh ID
NDC	x	y	z	w

Training & Evaluation

All training and hyperparameter settings are in setting.py.

Train SSR

python3 train.py

Test with different best checkpints

python3 test.py --checkpoint psnr
python3 test.py --checkpoint ssim
python3 test.py --checkpoint rmse

Baselines

We additionally provide baselines reproduction code:

Monte Carlo Denoising via Auxiliary Feature Guided Self-Attention (TOG 2021)

Interactive Monte Carlo Denoising using Affinity of Neural Features (SIGGRAPH 2021)

Neural Supersampling for Real-time Rendering (SIGGRAPH 2020)

Interactive Reconstruction of Monte Carlo Image Sequences using a Recurrent Denoising Autoencoder (SIGGRAPH 2017)

Citing

@article{zhang2023high,
  title={High-Quality Real-Time Rendering Using Subpixel Sampling Reconstruction},
  author={Zhang, Boyu and Yuan, Hongliang and Zhu, Mingyan and Liu, Ligang and Wang, Jue},
  journal={arXiv preprint arXiv:2301.01036v2},
  year={2023}
}

or

@article{zhang2023high,
  title={High-Quality Supersampling via Mask-reinforced Deep Learning for Real-time Rendering},
  author={Zhang, Boyu and Yuan, Hongliang and Zhu, Mingyan and Liu, Ligang and Wang, Jue},
  journal={arXiv preprint arXiv:2301.01036v1},
  year={2023}
}

Luciferbobo / SSR