In this paper, we propose an algorithm to estimate missing values in tensors of visual data. The can be due problems the acquisition process or because user manually identified unwanted outliers. Our works even with a small amount samples and it propagate structure fill larger regions. methodology is built on recent studies about matrix completion using trace norm. contribution our paper extend case tensor by proposing first definition norm for then building working algorithm. First, that...

We propose an efficient algorithm to embed a given image into the latent space of StyleGAN. This embedding enables semantic editing operations that can be applied existing photographs. Taking StyleGAN trained on FFHD dataset as example, we show results for morphing, style transfer, and expression transfer. Studying provides valuable insights structure space. set experiments test what class images embedded, how they are is suitable embedding, if semantically meaningful.

CGA shape , a novel grammar for the procedural modeling of CG architecture, produces building shells with high visual quality and geometric detail. It extensive architectural models computer games movies, at low cost. Context sensitive rules allow user to specify interactions between entities hierarchical descriptions. Selected examples demonstrate solutions previously unsolved problems, especially consistent mass volumetric shapes arbitrary orientation. is shown efficiently generate massive...

We propose Image2StyleGAN++, a flexible image editing framework with many applications. Our extends the recent Image2StyleGAN in three ways. First, we introduce noise optimization as complement to W+ latent space embedding. can restore high frequency features images and thus significantly improves quality of reconstructed images, e.g. big increase PSNR from 20 dB 45 dB. Second, extend global embedding enable local embeddings. Third, combine activation tensor manipulation perform edits along...

This paper describes algorithms to automatically derive 3D models of high visual quality from single facade images arbitrary resolutions. We combine the procedural modeling pipeline shape grammars with image analysis a meaningful hierarchical subdivision. Our system gives rise three exciting applications: urban reconstruction based on low resolution oblique aerial imagery, facades higher ground-based and automatic derivation grammar rules build rule base for technology.

Abstract This paper provides a comprehensive overview of urban reconstruction. While there exists considerable body literature, this topic is still under active research. The work reviewed in survey stems from the following three research communities: computer graphics, vision and photogrammetry remote sensing. Our goal to provide that will help researchers better position their own context existing solutions, newcomers practitioners graphics quickly gain an vast field. Further, we would...

This paper presents a new method for the automatic modeling of architecture. Building designs are derived using split grammars, type parametric set grammar based on concept shape. The also introduces an attribute matching system and separate control grammar, which offer flexibility required to model buildings large variety different styles design ideas. Through adaptive nature used, created building can either be generic or adhere closely specified goal, depending amount data available.

We address the problem of estimating a high quality dense depth map from single RGB input image. start out with baseline encoder-decoder convolutional neural network architecture and pose question how global processing information can help improve overall estimation. To this end, we propose transformer-based block that divides range into bins whose center value is estimated adaptively per The final values are as linear combinations bin centers. call our new building AdaBins. Our results show...

We propose semantic region-adaptive normalization (SEAN), a simple but effective building block for Generative Adversarial Networks conditioned on segmentation masks that describe the regions in desired output image. Using SEAN normalization, we can build network architecture control style of each region individually, e.g., specify one reference image per region. is better suited to encode, transfer, and synthesize than best previous method terms reconstruction quality, variability, visual...

In this paper, a new two-step approach (detecting and pruning) for automatic extraction of road networks from aerial images is presented. The detection step based on shape classification local homogeneous region around pixel. enclosed by polygon, called the footprint This involves detecting footprints, tracking roads, growing tree. We use spoke wheel operator to obtain footprint. propose an seeding method rectangular approximations footprints toe-finding algorithm classify tree pruning makes...

Accurate depth estimation from images is a fundamental task in many applications including scene understanding and reconstruction. Existing solutions for often produce blurry approximations of low resolution. This paper presents convolutional neural network computing high-resolution map given single RGB image with the help transfer learning. Following standard encoder-decoder architecture, we leverage features extracted using high performing pre-trained networks when initializing our encoder...

The ability to generate novel, diverse, and realistic 3D shapes along with associated part semantics structure is central many applications requiring high-quality assets or large volumes of training data. A key challenge towards this goal how accommodate diverse shape variations, including both continuous deformations parts as well structural discrete alterations which add to, remove from, modify the constituents compositional structure. Such object can typically be organized into a...

We propose a novel framework for reconstructing lightweight polygonal surfaces from point clouds. Unlike traditional methods that focus on either extracting good geometric primitives or obtaining proper arrangements of primitives, the emphasis this work lies in intersecting (planes only) and seeking an appropriate combination them to obtain manifold surface model without boundary. show reconstruction clouds can be cast as binary labeling problem. Our method is based hypothesizing selection...

High-quality, diverse, and photorealistic images can now be generated by unconditional GANs (e.g., StyleGAN). However, limited options exist to control the generation process using (semantic) attributes, while still preserving quality of output. Further, due entangled nature GAN latent space, performing edits along one attribute easily result in unwanted changes other attributes. In this paper, context conditional exploration spaces, we investigate two sub-problems attribute-conditioned...

This paper tackles the problem of depth estimation from a single image. Existing work either focuses on generalization performance disregarding metric scale, i.e. relative estimation, or state-of-the-art results specific datasets, estimation. We propose first approach that combines both worlds, leading to model with excellent while maintaining scale. Our flagship model, ZoeD-M12-NK, is pre-trained 12 datasets using and fine-tuned two depth. use lightweight head novel bin adjustment design...

We introduce 3DShape2VecSet, a novel shape representation for neural fields designed generative diffusion models. Our can encode 3D shapes given as surface models or point clouds, and represents them fields. The concept of has previously been combined with global latent vector, regular grid vectors, an irregular vectors. new encodes on top set draw from multiple concepts, such the radial basis function representation, cross attention self-attention function, to design learnable that is...

We present Magic123, a two-stage coarse-to-fine approach for high-quality, textured 3D meshes generation from single unposed image in the wild using both2D and priors. In first stage, we optimize neural radiance field to produce coarse geometry. second adopt memory-efficient differentiable mesh representation yield high-resolution with visually appealing texture. both stages, content is learned through reference view supervision novel views guided by combination of 2D diffusion introduce...

Abstract The field of visual computing is rapidly advancing due to the emergence generative artificial intelligence (AI), which unlocks unprecedented capabilities for generation, editing, and reconstruction images, videos, 3D scenes. In these domains, diffusion models are AI architecture choice. Within last year alone, literature on diffusion‐based tools applications has seen exponential growth relevant papers published across computer graphics, vision, communities with new works appearing...

This paper presents a new method for the automatic modeling of architecture. Building designs are derived using split grammars, type parametric set grammar based on concept shape. The also introduces an attribute matching system and separate control grammar, which offer flexibility required to model buildings large variety different styles design ideas. Through adaptive nature used, created building can either be generic or adhere closely specified goal, depending amount data available.

This paper describes algorithms to automatically derive 3D models of high visual quality from single facade images arbitrary resolutions. We combine the procedural modeling pipeline shape grammars with image analysis a meaningful hierarchical subdivision. Our system gives rise three exciting applications: urban reconstruction based on low resolution oblique aerial imagery, facades higher ground-based and automatic derivation grammar rules build rule base for technology.

CGA shape, a novel shape grammar for the procedural modeling of CG architecture, produces building shells with high visual quality and geometric detail. It extensive architectural models computer games movies, at low cost. Context sensitive rules allow user to specify interactions between entities hierarchical descriptions. Selected examples demonstrate solutions previously unsolved problems, especially consistent mass volumetric shapes arbitrary orientation. is shown efficiently generate...

This sketch presents a solution to efficiently model the street networks of large urban areas. Parish and Müller [2001] were first note that network is key create model. While this algorithm created high quality solution, method does not allow incorporate user-control. To address limitation we provide rather different alternative modeling allows integrate wide variety user input. The idea use tensor fields guide generation graphs. A can interactively edit graph by either modifying underlying...

Automatically generating 3D building models from 2D architectural drawings has many useful applications in the architecture engineering and construction community. This survey of model generation paper CAD-based covers common pipeline compares various algorithms for each step process.

We introduce a real-time interactive visual editing paradigm for shape grammars, allowing the creation of rulebases from scratch without text file editing. In previous work, shape-grammar based procedural techniques were successfully applied to architectural models. However, those methods are based, and may therefore be difficult use artists with little computer science background. Therefore goal was enable work-flow combining power grammars traditional modeling techniques. extend grammar...