Graphical Models最新文献

Asset creation in game development usually requires multiple iterations until a final version is achieved. This iterative process becomes more significant when the content is pixel art, in which the artist carefully places each pixel. We hypothesize that the problem of generating character sprites in a target pose (e.g., facing right) given a source (e.g., facing front) can be framed as an image-to-image translation task. Then, we present an architecture of deep generative models that takes as input an image of a character in one domain (pose) and transfers it to another. We approach the problem using generative adversarial networks (GANs) and build on Pix2Pix’s architecture while leveraging some specific characteristics of the pixel art style. We evaluated the trained models using four small datasets (less than 1k) and a more extensive and diverse one (12k). The models yielded promising results, and their generalization capacity varies according to the dataset size and variability. After training models to generate images among four domains (i.e., front, right, back, left), we present an early version of a mixed-initiative sprite editor that allows users to interact with them and iterate in creating character sprites.

Crowd simulation is a research area widely used in diverse fields, including gaming and security, assessing virtual agent movements through metrics like time to reach their goals, speed, trajectories, and densities. This is relevant for security applications, for instance, as different crowd configurations can determine the time people spend in environments trying to evacuate them. In this work, we extend WebCrowds, an authoring tool for crowd simulation, to allow users to build scenarios and evaluate them through a set of metrics. The aim is to provide a quantitative metric that can, based on simulation data, select the best crowd configuration in a certain environment. We conduct experiments to validate our proposed metric in multiple crowd simulation scenarios and perform a comparison with another metric found in the literature. The results show that experts in the domain of crowd scenarios agree with our proposed quantitative metric.

This paper presents performance improvements for Ellipsoid Clipmaps, an out-of-core planet-sized geodetically accurate terrain rendering algorithm. The performance improvements were achieved by eliminating unnecessarily dense levels, more accurate block culling in the geographic coordinate system, and more efficient rendering methods. The elimination of unnecessarily dense levels is the result of analyzing and determining the optimal relative height of the viewer with respect to the most detailed level, resulting in the most consistent size of triangles across all visible levels. The proposed method for estimating the visibility of blocks based on view orientation allows rapid block-level view frustum culling performed in data space before visualization and spatial transformation of blocks. The use of a modern geometry pipeline through task and mesh shaders forced the handling of extremely fine granularity of blocks, but also shifted a significant part of the block culling process from CPU to the GPU. The approach described achieves high throughput and enables geodetically accurate rendering of the terrain based on the WGS 84 reference ellipsoid at very high resolution and in real time, with tens of millions of triangles with an average area of about 0.5 pix${}^2$ on a 1080p screen on mid-range graphics cards.

This paper aims at extending the method of Zhang et al. (2023) to produce not only portrait bas-reliefs from single photographs, but also high-depth reliefs with reasonable depth ordering. We cast this task as a problem of style-aware photo-to-depth translation, where the input is a photograph conditioned by a style vector and the output is a portrait relief with desired depth style. To construct ground-truth data for network training, we first propose an optimization-based method to synthesize high-depth reliefs from 3D portraits. Then, we train a normal-to-depth network to learn the mapping from normal maps to relief depths. After that, we use the trained network to generate high-depth relief samples using the provided normal maps from Zhang et al. (2023). As each normal map has pixel-wise photograph, we are able to establish correspondences between photographs and high-depth reliefs. By taking the bas-reliefs of Zhang et al. (2023), the new high-depth reliefs and their mixtures as target ground-truths, we finally train a encoder-to-decoder network to achieve style-aware relief modeling. Specially, the network is based on a U-shaped architecture, consisting of Swin Transformer blocks to process hierarchical deep features. Extensive experiments have demonstrated the effectiveness of the proposed method. Comparisons with previous works have verified its flexibility and state-of-the-art performance.

Modern laser scanners, depth sensor devices and Dense Image Matching techniques allow for capturing of extensive point cloud datasets. While capturing has become more user-friendly, the size of registered point clouds results in large datasets which pose challenges for processing, storage and visualization. This paper presents a decomposition scheme using oriented KD trees and the wavelet transform for unordered point clouds. Taking inspiration from image pyramids, the decomposition scheme comes with a Level of Detail representation where higher-levels are progressively reconstructed from lower ones, thus making it suitable for streaming and continuous Level of Detail. Furthermore, the decomposed representation allows common compression techniques to achieve higher compression ratios by modifying the underlying frequency data at the cost of geometric accuracy and therefore allows for flexible lossy compression. After introducing this novel decomposition scheme, results are discussed to show how it deals with data captured from different sources.

Reconstructing 3D human pose and body shape from monocular images or videos is a fundamental task for comprehending human dynamics. Frame-based methods can be broadly categorized into two fashions: those regressing parametric model parameters (e.g., SMPL) and those exploring alternative representations (e.g., volumetric shapes, 3D coordinates). Non-parametric representations have demonstrated superior performance due to their enhanced flexibility. However, when applied to video data, these non-parametric frame-based methods tend to generate inconsistent and unsmooth results. To this end, we present a novel approach that directly regresses the 3D coordinates of the mesh vertices and body joints with a spatial–temporal Transformer. In our method, we introduce a SpatioTemporal Learning Block (STLB) with Spatial Learning Module (SLM) and Temporal Learning Module (TLM), which leverages spatial and temporal information to model interactions at a finer granularity, specifically at the body token level. Our method outperforms previous state-of-the-art approaches on Human3.6M and 3DPW benchmark datasets.

Image enhancement is an indispensable pre-processing step for several image processing applications. Mainly, histogram equalization is one of the widespread techniques used by various researchers to improve the image quality by expanding the pixel values to fill the entire dynamic grayscale. It results in the visual artifact, structural information loss near edges due to the information loss (due to many-to-one mapping), and alteration in average luminance to a higher value. This paper proposes an enhancement algorithm based on structural information for homogeneous background images. The intensities are divided into two segments using the median value to preserve the average luminance. Unlike traditional techniques, this algorithm incorporates the spatial locations in the equalization process instead of the number of intensity values occurrences. The occurrences of each intensity concerning their spatial locations are combined using Rènyi entropy to enumerate a discrete function. An adaptive clipping limit is applied to the discrete function to control the enhancement rate. Then histogram equalization is performed on each segment separately, and the equalized segments are integrated to produce an enhanced image. The algorithm’s effectiveness is validated by evaluating the proposed method on CEED, CSIQ, LOL, and TID2013 databases. Experimental results reveal that the proposed method improves the contrast while preserving structural information, detail information, and average luminance. They are quantified by the high value of contrast improvement index, structural similarity index, and discrete entropy, and low value of average mean brightness error values of the proposed method when compared with the methods available in the literature, including deep learning architectures.

Differentiable rendering has been proven as a powerful tool to bridge 2D images and 3D models. With the aid of differentiable rendering, tasks in computer vision and computer graphics could be solved more elegantly and accurately. To address challenges in the implementations of differentiable rendering methods, we present an efficient and modular differentiable rendering library named Jrender based on Jittor. Jrender supports surface rendering for 3D meshes and volume rendering for 3D volumes. Compared with previous differentiable renderers, Jrender exhibits a significant improvement in both performance and rendering quality. Due to the modular design, various rendering effects such as PBR materials shading, ambient occlusions, soft shadows, global illumination, and subsurface scattering could be easily supported in Jrender, which are not available in other differentiable rendering libraries. To validate our library, we integrate Jrender into applications such as 3D object reconstruction and NeRF, which show that our implementations could achieve the same quality with higher performance.

Packing a designated set of shapes on a regular grid is an important class of operations research problems that has been intensively studied for more than six decades. Representing a $d$-dimensional discrete grid as $Z^d$, we formalise the generalised regular grid (GRG) as a surjective function from $Z^d$ to a geometric tessellation in a physical space, for example, the cube coordinates of a hexagonal grid or a quasilattice. This study employs 0-1 integer linear programming (ILP) to formulate the polyomino tiling problem with adjacency constraints. Rotation & reflection invariance in adjacency are considered. We separate the formal ILP from the topology & geometry of various grids, such as Ammann-Beenker tiling, Penrose tiling and periodic hypercube. Based on cutting-edge solvers, we reveal an intuitive correspondence between the integer program (a pattern of algebraic rules) and the computer codes. Models of packing problems in the GRG have wide applications in production system, facility layout planning, and architectural design. Two applications in planning high-rise residential apartments are illustrated.