Layered Light Field Reconstruction for Defocus Blur

Vaidyanathan, Karthikeyan; Munkberg, Jacob; Clarberg, Petrik; Salvi, Marco

doi:10.1145/2699647

Cited by 23 publications

(39 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Methods that retain the full light field (individual samples) such as [Lehtinen et al 2011;Lehtinen et al 2012;Sen and Darabi 2012] can filter the noisy irradiance separately but have a high storage and reconstruction overhead. The concurrent work of [Vaidyanathan et al 2014] demonstrates a much faster, interactive formulation of sheared filtering for depth of field. They separate the image into depth layers, and simplify the 4D filter into splatting and screen-space convolution steps.…”

Section: Previous Workmentioning

confidence: 99%

See 1 more Smart Citation

Factored axis-aligned filtering for rendering multiple distribution effects

et al. 2014

View full text Add to dashboard Cite

a) Our Method, 138 rpp, 3.14 sec (b) Equal time, 176 rpp, 3.11 sec (c) Our method, 138 rpp, 3.14 sec (d) Eq. quality, 5390 rpp, 130 sec (e) No factoring, 138 rpp, 3.10 sec defocus filter (pixels) num. primary rays (f) Primary filter and rpp 1 20 1 64 indirect filter (pixels) num. indirect rays (g) Indirect filter and rpp Figure 1: (a) The CHESS scene, with defocus blur, area light direct and indirect illumination, rendered at 900×1024 with an average 138 atomic rays per pixel (rpp), in 3.14 sec on an NVIDIA Titan GPU. We compare to different methods in the insets. Readers are encouraged to zoom into the PDF to examine the noise. The top row inset is a sharp in-focus region while the other two regions are defocus blurred; all insets include noisy direct and indirect illumination. In (b) we compare to equal time stratified Monte Carlo (MC) sampling with 176 rpp; in (c), Our method; and in (d), Equal quality MC with 5390 rpp is 40× slower. One of the key contributions of our method is factoring of texture and irradiance, so that irradiance can be pre-filtered before combining with texture. Without factoring, the defocus filter cannot remove noise for in-focus regions as shown in (e), top inset. In (f) and (g) top row, we show filter size for texture and indirect irradiance. Black pixels indicate that factoring cannot be used and irradiance cannot be pre-filtered. In the bottom row we show number of primary rays and indirect samples respectively. Our method uses separate sampling rates and filters for primary and secondary effects which makes it more effective. AbstractMonte Carlo (MC) ray-tracing for photo-realistic rendering often requires hours to render a single image due to the large sampling rates needed for convergence. Previous methods have attempted to filter sparsely sampled MC renders but these methods have high reconstruction overheads. Recent work has shown fast performance for individual effects, like soft shadows and indirect illumination, using axis-aligned filtering. While some components of light transport such as indirect or area illumination are smooth, they are often multiplied by high-frequency components such as texture, which prevents their sparse sampling and reconstruction.We propose an approach to adaptively sample and filter for simultaneously rendering primary (defocus blur) and secondary (soft shadows and indirect illumination) distribution effects, based on a multi-dimensional frequency analysis of the direct and indirect illumination light fields. We describe a novel approach of factoring texture and irradiance in the presence of defocus blur, which allows for pre-filtering noisy irradiance when the texture is not noisy. Our approach naturally allows for different sampling rates for primary and secondary effects, further reducing the overall ray count. While the theory considers only Lambertian surfaces, we obtain promising results for moderately glossy surfaces. We demonstrate 30× sampling rate reduction compared to equal quality noise-free MC. Combined with a GPU implementation and l...

show abstract

Section: Previous Workmentioning

confidence: 99%

“…Neither of these papers explicitly equates the slope of the defocused light field to the local circles of confusion. The concurrent work of [Vaidyanathan et al 2014] does derive a very similar frequency analysis for defocus blur, but does not explore the connections with direct and indirect illumination that we study next.…”

Section: Defocus Blurmentioning

confidence: 99%

Factored axis-aligned filtering for rendering multiple distribution effects

et al. 2014

View full text Add to dashboard Cite

show abstract

“…However, this algorithm was reported to be 3 – 4× slower than their previous method. Vaidyanathan et al [VMCS13] reconstruct defocus blur by partitioning samples in screen space tiles and depth layers, and derive an anisotropic, separable reconstruction filter for each partition. In the combined case of defocus and motion blur, the light field is anisotropic in a five‐dimensional space, and their reconstruction filter is no longer separable.…”

Section: Previous Workmentioning

confidence: 99%

“…This, somewhat coarser, approximation allows for a very efficient implementation. We present a non‐trivial extension of the filter framework of Vaidyanathan et al [VMCS13], and show that the 5D filter is separable in xut and yv . We carefully design a novel warped Gaussian distribution such that even an anisotropic xut filter can be plugged into their framework.…”

Section: Overviewmentioning

confidence: 99%

See 1 more Smart Citation

Layered Reconstruction for Defocus and Motion Blur

Munkberg

Vaidyanathan

Hasselgren

et al. 2014

Computer Graphics Forum

Self Cite

View full text Add to dashboard Cite

Input samples (8 spp) Our filter 32 × 32 Our filter 44 × 44 Reference (1024 spp) 80 ms 436 ms Figure 1: The San Miguel scene with depth of field and large camera motion. The closeups show the input samples, the result using our new reconstruction algorithm on the input samples, and the reference rendered with 1024 spp. The two images for our algorithm show different filter search window configurations, where a larger window improves quality at the expense of performance. Our novel reconstruction method dramatically reduces noise, and produces images close to the reference. AbstractLight field reconstruction algorithms can substantially decrease the noise in stochastically rendered images. Recent algorithms for defocus blur alone are both fast and accurate. However, motion blur is a considerably more complex type of camera effect, and as a consequence, current algorithms are either slow or too imprecise to use in high quality rendering. We extend previous work on real-time light field reconstruction for defocus blur to handle the case of simultaneous defocus and motion blur. By carefully introducing a few approximations, we derive a very efficient sheared reconstruction filter, which produces high quality images even for a low number of input samples. Our algorithm is temporally robust, and is about two orders of magnitude faster than previous work, making it suitable for both real-time rendering and as a post-processing pass for offline rendering.

show abstract

Spectral analysis of a surface occlusion model for image-based rendering sampling

Chen

Zhu²

2022

Digital Signal Processing

View full text Add to dashboard Cite

Layered Light Field Reconstruction for Defocus Blur

Cited by 23 publications

References 29 publications

Factored axis-aligned filtering for rendering multiple distribution effects

Factored axis-aligned filtering for rendering multiple distribution effects

Layered Reconstruction for Defocus and Motion Blur

Spectral analysis of a surface occlusion model for image-based rendering sampling

Contact Info

Product

Resources

About