Rendering 2021 - DL-only Track

Permanent URI for this collection

https://diglib.eg.org/handle/10.2312/2633075

Browse

Browsing Rendering 2021 - DL-only Track by Subject "Image"

Now showing 1 - 4 of 4
  • Thumbnail Image
    Item
    Fast Polygonal Splatting using Directional Kernel Difference
    (The Eurographics Association, 2021) Moroto, Yuji; Hachisuka, Toshiya; Umetani, Nobuyuki; Bousseau, Adrien and McGuire, Morgan
    Depth-of-field (DoF) filtering is an important image-processing task for producing blurred images similar to those obtained with a large aperture camera lens. DoF filtering applies an image convolution with a spatially varying kernel and is thus computationally expensive, even on modern computational hardware. In this paper, we introduce an approach for fast and accurate DoF filtering for polygonal kernels, where the value is constant inside the kernel. Our approach is an extension of the existing approach based on discrete differenced kernels. The performance gain here hinges upon the fact that kernels typically become sparse (i.e., mostly zero) when taking the difference. We extended the existing approach to conventional axis-aligned differences to non-axis-aligned differences. The key insight is that taking such differences along the directions of the edges makes polygonal kernels significantly sparser than just taking the difference along the axis-aligned directions, as in existing studies. Compared to a naive image convolution, we achieve an order of magnitude speedup, allowing a real-time application of polygonal kernels even on high-resolution images.
  • Thumbnail Image
    Item
    NeLF: Neural Light-transport Field for Portrait View Synthesis and Relighting
    (The Eurographics Association, 2021) Sun, Tiancheng; Lin, Kai-En; Bi, Sai; Xu, Zexiang; Ramamoorthi, Ravi; Bousseau, Adrien and McGuire, Morgan
    Human portraits exhibit various appearances when observed from different views under different lighting conditions. We can easily imagine how the face will look like in another setup, but computer algorithms still fail on this problem given limited observations. To this end, we present a system for portrait view synthesis and relighting: given multiple portraits, we use a neural network to predict the light-transport field in 3D space, and from the predicted Neural Light-transport Field (NeLF) produce a portrait from a new camera view under a new environmental lighting. Our system is trained on a large number of synthetic models, and can generalize to different synthetic and real portraits under various lighting conditions. Our method achieves simultaneous view synthesis and relighting given multi-view portraits as the input, and achieves state-of-the-art results.
  • Thumbnail Image
    Item
    Sampling Clear Sky Models using Truncated Gaussian Mixtures
    (The Eurographics Association, 2021) Vitsas, Nick; Vardis, Konstantinos; Papaioannou, Georgios; Bousseau, Adrien and McGuire, Morgan
    Parametric clear sky models are often represented by simple analytic expressions that can efficiently generate plausible, natural radiance maps of the sky, taking into account expensive and hard to simulate atmospheric phenomena. In this work, we show how such models can be complemented by an equally simple, elegant and generic analytic continuous probability density function (PDF) that provides a very good approximation to the radiance-based distribution of the sky. We describe a fitting process that is used to properly parameterise a truncated Gaussian mixture model, which allows for exact, constant-time and minimal-memory sampling and evaluation of this PDF, without rejection sampling, an important property for practical applications in offline and real-time rendering. We present experiments in a standard importance sampling framework that showcase variance reduction approaching that of a more expensive inversion sampling method using Summed Area Tables.
  • Thumbnail Image
    Item
    Single-image Full-body Human Relighting
    (The Eurographics Association, 2021) Lagunas, Manuel; Sun, Xin; Yang, Jimei; Villegas, Ruben; Zhang, Jianming; Shu, Zhixin; Masia, Belen; Gutierrez, Diego; Bousseau, Adrien and McGuire, Morgan
    We present a single-image data-driven method to automatically relight images with full-body humans in them. Our framework is based on a realistic scene decomposition leveraging precomputed radiance transfer (PRT) and spherical harmonics (SH) lighting. In contrast to previous work, we lift the assumptions on Lambertian materials and explicitly model diffuse and specular reflectance in our data. Moreover, we introduce an additional light-dependent residual term that accounts for errors in the PRTbased image reconstruction. We propose a new deep learning architecture, tailored to the decomposition performed in PRT, that is trained using a combination of L1, logarithmic, and rendering losses. Our model outperforms the state of the art for full-body human relighting both with synthetic images and photographs.