We present a practical and robust deep learning solution for capturing rendering novel views of complex real world scenes virtual exploration. Previous approaches either require intractably dense view sampling or provide little to no guidance how users should sample scene reliably render high-quality views. Instead, we propose an algorithm synthesis from irregular grid sampled that first expands each into local light field via multiplane image (MPI) representation, then renders by blending...

With the introduction of consumer light field cameras, imaging has recently become widespread. However, there is an inherent trade-off between angular and spatial resolution, thus, these cameras often sparsely sample in either or domain. In this paper, we use machine learning to mitigate trade-off. Specifically, propose a novel learning-based approach synthesize new views from sparse set input views. We build upon existing view synthesis techniques break down process into disparity color...

Producing a high dynamic range (HDR) image from set of images with different exposures is challenging process for scenes. A category existing techniques first register the input to reference and then merge aligned into an HDR image. However, artifacts registration usually appear as ghosting tearing in final images. In this paper, we propose learning-based approach address problem We use convolutional neural network (CNN) our learning model present compare three system architectures process....

High dynamic range (HDR) imaging from a set of sequential exposures is an easy way to capture high-quality images static scenes, but suffers artifacts for scenes with significant motion. In this paper, we propose new approach HDR reconstruction that draws information all the more robust camera/scene motion than previous techniques. Our algorithm based on novel patch-based energy-minimization formulation integrates alignment and in joint optimization through equation call image synthesis...

The most successful approaches for filtering Monte Carlo noise use feature-based filters (e.g., cross-bilateral and cross non-local means filters) that exploit additional scene features such as world positions shading normals. However, their main challenge is finding the optimal weights each feature in filter to reduce but preserve detail. In this paper, we observe there a complex relationship between noisy data ideal parameters, propose learn using nonlinear regression model. To do this,...

Digital cameras can only capture a limited range of real-world scenes' luminance, producing images with saturated pixels. Existing single image high dynamic (HDR) reconstruction methods attempt to expand the but are not able hallucinate plausible textures, results artifacts in areas. In this paper, we present novel learning-based approach reconstruct an HDR by recovering pixels input LDR visually pleasing way. Previous deep apply same convolutional filters on well-exposed and pixels,...

Light field cameras have many advantages over traditional cameras, as they allow the user to change various camera settings after capture. However, capturing light fields requires a huge bandwidth record data: modern can only take three images per second. This prevents current consumer from videos. Temporal interpolation at such extreme scale (10x, 3 fps 30 fps) is infeasible too much information will be entirely missing between adjacent frames. Instead, we develop hybrid imaging system,...

Despite significant progress in high dynamic range (HDR) imaging over the years, it is still difficult to capture high-quality HDR video with a conventional, off-the-shelf camera. The most practical way do this alternating exposures for every LDR frame and then use an alignment method based on optical flow register together. However, results objectionable artifacts whenever there complex motion fails. To address problem, we propose new approach reconstruction from exposure sequences that...

Image-based texture mapping is a common way of producing maps for geometric models real-world objects. Although high-quality map can be easily computed accurate geometry and calibrated cameras, the quality degrades significantly in presence inaccuracies. In this paper, we address problem by proposing novel global patch-based optimization system to synthesize aligned images. Specifically, use synthesis reconstruct set photometrically-consistent images drawing information from source Our...

We present a practical and robust deep learning solution for capturing rendering novel views of complex real world scenes virtual exploration. Previous approaches either require intractably dense view sampling or provide little to no guidance how users should sample scene reliably render high-quality views. Instead, we propose an algorithm synthesis from irregular grid sampled that first expands each into local light field via multiplane image (MPI) representation, then renders by blending...

This paper presents a Multiplicative Patchwork Method (MPM) for audio watermarking. The watermark signal is embedded by selecting two subsets of the host features and modifying one subset multiplicatively regarding data, whereas another left unchanged. method implemented in wavelet domain approximation coefficients are used to embed data. In order have an error-free detection, data inserted only frames where ratio energy between predefined values. Also control inaudibility insertion, we use...

Abstract A practical way to generate a high dynamic range (HDR) video using off‐the‐shelf cameras is capture sequence with alternating exposures and reconstruct the missing content at each frame. Unfortunately, existing approaches are typically slow not able handle challenging cases. In this paper, we propose learning‐based approach address difficult problem. To do this, use two sequential convolutional neural networks (CNN) model entire HDR reconstruction process. first step, align...

A robust image watermarking scheme in the ridgelet transform domain is proposed this paper. Due to use of domain, sparse representation an which deals with line singularities obtained. In order achieve more robustness and transparency, watermark data embedded selected blocks host by modifying amplitude coefficients represent most energetic direction. Since probability distribution function not known, we propose a universally optimum decoder perform extraction distribution-independent...

In this paper, a novel arrangement for quantizer levels in the Quantization Index Modulation (QIM) method is proposed. Due to perceptual advantages of logarithmic quantization, and order solve problems previous quantization-based method, we used compression function <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">¿</i> -Law standard quantization. regard, host signal first transformed into domain using function. Then, data quantized uniformly...

Abstract In this paper, we propose a deep learning approach for estimating the spatially‐varying BRDFs (SVBRDF) from single image. Most existing techniques use pixel‐wise loss functions which limits flexibility of networks in handling highly unconstrained problem. Moreover, since obtaining ground truth SVBRDF parameters is difficult, most methods typically train their on synthetic images and, therefore, do not effectively generalize to real examples. To avoid these limitations, an...

We propose a learning-based approach to synthesize light field with small baseline from single image. the novel view images by first using convolutional neural network (CNN) promote input image into layered representation of scene. extend multiplane (MPI) allowing disparity layers be inferred show that, compared original MPI representation, our models scenes more accurately. Moreover, we handle visible and occluded regions separately through two parallel networks. The synthesized these...

Recent methods (e.g. MaterialGAN) have used unconditional GANs to generate per-pixel material maps, or as a prior reconstruct materials from input photographs. These models can varied random appearance, but do not any mechanism constrain the generated specific category control coarse structure of material, such exact brick layout on wall. Furthermore, reconstructed single photo commonly artifacts and are generally tileable, which limits their use in practical content creation pipelines. We...

Authoring high-quality digital materials is key to realism in 3D rendering. Previous generative models for have been trained exclusively on synthetic data; such data limited availability and has a visual gap real materials. We circumvent this limitation by proposing PhotoMat: the first material generator photos of samples captured using cell phone camera with flash. Supervision individual maps not available setting. Instead, we train neural representation that rendered learned relighting...

Abstract Monte Carlo rendering systems can produce important visual effects such as depth of field, motion blur, and area lighting, but the rendered images suffer from objectionable noise at low sampling rates. Although years research in image processing has produced powerful denoising algorithms, most them assume that is spatially‐invariant over entire cannot be directly applied to denoise rendering. In this paper, we propose a new approach enables use any technique remove renderings. Our...

Rendering specular material appearance is a core problem of computer graphics. While smooth analytical models are widely used, the high-frequency structure real highlights requires considering discrete, finite microgeometry. Instead explicit modeling and simulation surface microstructure (which was explored in previous work), we propose novel direction: learning directional patterns from synthetic or measured examples, by training generative adversarial network (GAN). A key challenge...

Slow motion videos are becoming increasingly popular, but capturing high-resolution at extremely high frame rates requires professional high-speed cameras. To mitigate this problem, current techniques increase the rate of standard through interpolation by assuming linear object which is not valid in challenging cases. In paper, we address problem using two video streams as input; an auxiliary with and low spatial resolution, providing temporal information, addition to main resolution. We...

Abstract Recently, deep learning approaches have proven successful at removing noise from Monte Carlo (MC) rendered images extremely low sampling rates, e.g., 1–4 samples per pixel (spp). While these methods provide dramatic speedups, they operate on uniformly sampled MC images. However, the full promise of sample counts requires both adaptive and reconstruction/denoising. Unfortunately, traditional techniques fail to handle cases with since there is insufficient information reliably...

In this paper, we propose a novel optimization-based method to estimate the reflectance properties of near planar surface from single input image. Specifically, perform test-time optimization by directly updating parameters neural network minimize test error. Since image SVBRDF estimation is highly ill-posed problem, such an prone overfitting. Our main contribution address problem introducing training mechanism that takes into account. train our minimizing loss after one or more gradient...

In this paper, we propose an algorithm to interpolate between a pair of images dynamic scene. While in the past years significant progress frame interpolation has been made, current approaches are not able handle with brightness and illumination changes, which common even when captured shortly apart. We address problem by taking advantage existing optical flow methods that highly robust variations illumination. Specifically, using bidirectional flows estimated pre-trained network, predict...