-
Decorrelating ReSTIR Samplers via MCMC Mutations
Authors:
Rohan Sawhney,
Daqi Lin,
Markus Kettunen,
Benedikt Bitterli,
Ravi Ramamoorthi,
Chris Wyman,
Matt Pharr
Abstract:
Monte Carlo rendering algorithms often utilize correlations between pixels to improve efficiency and enhance image quality. For real-time applications in particular, repeated reservoir resampling offers a powerful framework to reuse samples both spatially in an image and temporally across multiple frames. While such techniques achieve equal-error up to 100 times faster for real-time direct lightin…
▽ More
Monte Carlo rendering algorithms often utilize correlations between pixels to improve efficiency and enhance image quality. For real-time applications in particular, repeated reservoir resampling offers a powerful framework to reuse samples both spatially in an image and temporally across multiple frames. While such techniques achieve equal-error up to 100 times faster for real-time direct lighting and global illumination, they are still far from optimal. For instance, unchecked spatiotemporal resampling often introduces noticeable correlation artifacts, while reservoirs holding more than one sample suffer from impoverishment in the form of duplicate samples. We demonstrate how interleaving Markov Chain Monte Carlo (MCMC) mutations with reservoir resampling helps alleviate these issues, especially in scenes with glossy materials and difficult-to-sample lighting. Moreover, our approach does not introduce any bias, and in practice we find considerable improvement in image quality with just a single mutation per reservoir sample in each frame.
△ Less
Submitted 31 October, 2022;
originally announced November 2022.
-
An unbiased ray-marching transmittance estimator
Authors:
Markus Kettunen,
Eugene d'Eon,
Jacopo Pantaleoni,
Jan Novak
Abstract:
We present an in-depth analysis of the sources of variance in state-of-the-art unbiased volumetric transmittance estimators, and propose several new methods for improving their efficiency. These combine to produce a single estimator that is universally optimal relative to prior work, with up to several orders of magnitude lower variance at the same cost, and has zero variance for any ray with non-…
▽ More
We present an in-depth analysis of the sources of variance in state-of-the-art unbiased volumetric transmittance estimators, and propose several new methods for improving their efficiency. These combine to produce a single estimator that is universally optimal relative to prior work, with up to several orders of magnitude lower variance at the same cost, and has zero variance for any ray with non-varying extinction. We first reduce the variance of truncated power-series estimators using a novel efficient application of U-statistics. We then greatly reduce the average expansion order of the power series and redistribute density evaluations to filter the optical depth estimates with an equidistant sampling comb. Combined with the use of an online control variate built from a sampled mean density estimate, the resulting estimator effectively performs ray marching most of the time while using rarely-sampled higher order terms to correct the bias.
△ Less
Submitted 20 February, 2021;
originally announced February 2021.
-
Behavioural Types for Memory and Method Safety in a Core Object-Oriented Language
Authors:
Mario Bravetti,
Adrian Francalanza,
Iaroslav Golovanov,
Hans Hüttel,
Mathias Steen Jakobsen,
Mikkel Klinke Kettunen,
António Ravara
Abstract:
We present a type-based analysis ensuring memory safety and object protocol completion in the Java-like language Mungo. Objects are annotated with usages, typestates-like specifications of the admissible sequences of method calls. The analysis entwines usage checking, controlling the order in which methods are called, with a static check determining whether references may contain null values. The…
▽ More
We present a type-based analysis ensuring memory safety and object protocol completion in the Java-like language Mungo. Objects are annotated with usages, typestates-like specifications of the admissible sequences of method calls. The analysis entwines usage checking, controlling the order in which methods are called, with a static check determining whether references may contain null values. The analysis prevents null pointer dereferencing and memory leaks and ensures that the intended usage protocol of every object is respected and completed. The type system has been implemented in the form of a type checker.
△ Less
Submitted 28 February, 2020;
originally announced February 2020.
-
E-LPIPS: Robust Perceptual Image Similarity via Random Transformation Ensembles
Authors:
Markus Kettunen,
Erik Härkönen,
Jaakko Lehtinen
Abstract:
It has been recently shown that the hidden variables of convolutional neural networks make for an efficient perceptual similarity metric that accurately predicts human judgment on relative image similarity assessment. First, we show that such learned perceptual similarity metrics (LPIPS) are susceptible to adversarial attacks that dramatically contradict human visual similarity judgment. While thi…
▽ More
It has been recently shown that the hidden variables of convolutional neural networks make for an efficient perceptual similarity metric that accurately predicts human judgment on relative image similarity assessment. First, we show that such learned perceptual similarity metrics (LPIPS) are susceptible to adversarial attacks that dramatically contradict human visual similarity judgment. While this is not surprising in light of neural networks' well-known weakness to adversarial perturbations, we proceed to show that self-ensembling with an infinite family of random transformations of the input --- a technique known not to render classification networks robust --- is enough to turn the metric robust against attack, while retaining predictive power on human judgments. Finally, we study the geometry imposed by our our novel self-ensembled metric (E-LPIPS) on the space of natural images. We find evidence of "perceptual convexity" by showing that convex combinations of similar-looking images retain appearance, and that discrete geodesics yield meaningful frame interpolation and texture morphing, all without explicit correspondences.
△ Less
Submitted 11 June, 2019; v1 submitted 10 June, 2019;
originally announced June 2019.
-
Context-Free Session Types for Applied Pi-Calculus
Authors:
Jens Aagaard,
Hans Hüttel,
Mathias Jakobsen,
Mikkel Kettunen
Abstract:
We present a binary session type system using context-free session types to a version of the applied pi-calculus of Abadi et. al. where only base terms, constants and channels can be sent. Session types resemble process terms from BPA and we use a version of bisimulation equivalence to characterize type equivalence. We present a quotiented type system defined on type equivalence classes for which…
▽ More
We present a binary session type system using context-free session types to a version of the applied pi-calculus of Abadi et. al. where only base terms, constants and channels can be sent. Session types resemble process terms from BPA and we use a version of bisimulation equivalence to characterize type equivalence. We present a quotiented type system defined on type equivalence classes for which type equivalence is built into the type system. Both type systems satisfy general soundness properties; this is established by an appeal to a generic session type system for psi-calculi.
△ Less
Submitted 26 August, 2018;
originally announced August 2018.
