Learning to predict scene depth from RGB inputs is a challenging task both for indoor and outdoor robot navigation. In this work we address unsupervised learning of ego-motion where supervision provided by monocular videos, as cameras are the cheapest, least restrictive most ubiquitous sensor robotics.
 Previous in image-to-depth has established strong baselines domain. We propose novel approach which produces higher quality results, able model moving objects shown transfer across data...

Building discriminative representations for 3D data has been an important task in computer graphics and vision research. Convolutional Neural Networks (CNNs) have shown to operate on 2D images with great success a variety of tasks. Lifting convolution operators (3DCNNs) seems like plausible promising next step. Unfortunately, the computational complexity CNNs grows cubically respect voxel resolution. Moreover, since most geometry are boundary based, occupied regions do not increase...

Abstract Procedural tree models have been popular in computer graphics for their ability to generate a variety of output trees from set input parameters and simulate plant interaction with the environment realistic placement virtual scenes. However, defining such is difficult task. We propose an inverse modelling approach stochastic that takes polygonal as estimates procedural model so it produces similar input. Our framework based on novel parametric generation uses Monte Carlo Markov...

We present an approach which takes advantage of both structure and semantics for unsupervised monocular learning depth ego-motion. More specifically we model the motions individual objects learn their 3D motion vector jointly with obtain more accurate results, especially challenging dynamic scenes not addressed by previous approaches. This is extended version Casser et al. [1]. Code models have been open sourced at: https://sites.google.com/corp/view/struct2depth.

Social navigation is the capability of an autonomous agent, such as a robot, to navigate in "socially compliant" manner presence other intelligent agents humans. With emergence autonomously navigating mobile robots human-populated environments (e.g., domestic service homes and restaurants food delivery on public sidewalks), incorporating socially compliant behaviors these becomes critical ensuring safe comfortable human-robot coexistence. To address this challenge, imitation learning...

We present a dynamic tree modeling and representation technique that allows complex models to interact with their environment. Our method uses changes in the light distribution proximity solid obstacles other trees as approximations of biologically motivated transformations on skeletal tree's main branches its procedurally generated foliage. Parts are transformed only when required, thus our approach is much faster than common algorithms such Open L-Systems or space colonization methods....

We introduce a novel framework for using natural language to generate and edit 3D indoor scenes, harnessing scene semantics text-scene grounding knowledge learned from large annotated databases. The advantage of editing interfaces is strongest when performing semantic operations at the sub-scene level, acting on groups objects. learn how manipulate these sub-scenes by analyzing existing scenes. perform edits first parsing command user transforming it into graph that used retrieve...

We introduce a novel method for reconstructing the 3D geometry of botanical trees from single photographs. Faithfully tree single-view sensor data is challenging and open problem because many possible exist that fit tree's shape observed view. address this challenge by defining reconstruction pipeline based on three neural networks. The networks simultaneously mask out in input photographs, identify species, obtain its radial bounding volume - our representation trees. Radial volumes (RBV)...

We present a lobe-based tree representation for modeling trees. The new is based on the observation that tree's foliage details can be abstracted into canonical geometry structures, termed lobe-textures. introduce techniques to (i) approximate of given data and encode it representation, (ii) decode synthesize fully detailed model visually resembles input. encoded serves as light intermediate which facilitates efficient storage transmission massive amounts trees, e.g., from server clients...

We present a lobe-based tree representation for modeling trees. The new is based on the observation that tree's foliage details can be abstracted into canonical geometry structures, termed lobe-textures. introduce techniques to (i) approximate of given data and encode it representation, (ii) decode synthesize fully detailed model visually resembles input. encoded serves as light intermediate which facilitates efficient storage transmission massive amounts trees, e.g., from server clients...

Due to the enormous amount of detail and interplay various biological phenomena, modeling realistic ecosystems trees other plants is a challenging open problem. Previous research on plant ecologies has focused representations handle this complexity, mostly through geometric simplifications, such as points or billboards. In paper we describe multi-scale method design large-scale with individual that are realistically modeled faithfully capture features, growth, interactions, different types...

Hands are dexterous and highly versatile manipulators that central to how humans interact with objects their environment. Consequently, modeling realistic hand-object interactions, including the subtle motion of individual fingers, is critical for applications in computer graphics, vision, mixed reality. Prior work on capturing interacting 3D focuses body object motion, often ignoring hand pose. In contrast, we introduce GRIP, a learning-based method takes, as input, object, synthesizes both...

We present a novel method for combining developmental tree models with turbulent wind fields. The geometry is created from internal growth functions of the model and its response to external stress induced by physically-plausible field that simulated Smoothed Particle Hydrodynamics (SPH). Our are dynamically evolving complex systems (1) react in real-time high-frequent changes simulation; (2) adapt long-term stress. extend this process wind-related effects such as branch breaking well bud...

Given a static tree model we present method to compute developmental stages that approximate the tree's natural growth. The is analyzed and graph-based description its skeleton determined. Based on structural similarity, branches are added where pruning has been applied or have died off over time. Botanic growth models allometric rules enable us produce convincing animations from young converge given model. Furthermore, user can explore all intermediate stages. By selectively applying...

Resulting from changing climatic conditions, wildfires have become an existential threat across various countries around the world. The complex dynamics paired with their often rapid progression renders disastrous natural phenomenon that is difficult to predict and counteract. In this paper we present a novel method for simulating goal realistically capture combustion process of individual trees resulting propagation fires at scale forests. We rely on state-of-the-art modeling approach...

Wildfires are a complex physical phenomenon that involves the combustion of variety flammable materials ranging from fallen leaves and dried twigs to decomposing organic material living flora. All these can potentially act as fuel with different properties determine progress severity wildfire. In this paper, we propose novel approach for simulating dynamic interaction between varying components wildfire, including processes convection, heat transfer vegetation, soil atmosphere. We...

Interactions play a key role in understanding objects and scenes for both virtual real-world agents. We introduce new general representation proximal interactions among physical that is agnostic to the type of or interaction involved. The based on tracking particles one participating then observing them with sensors appropriately placed volume surfaces. show how factorize these descriptors project into particular object so as obtain functional descriptor object, its landscape , capturing...

Real-life man-made objects often exhibit strong and easily-identifiable structure, as a direct result of their design or intended functionality. Structure typically appears in the form individual parts arrangement. Knowing about object structure can be an important cue for recognition scene understanding - key goal various AR robotics applications. However, commodity RGB-D sensors used these scenarios only produce raw, unorganized point clouds, without structural information captured scene....

We present a novel system for the interactive modeling of developmental climbing plants with an emphasis on efficient control and plausible physics response. A plant is represented by set connected anisotropic particles that respond to surrounding environment their inner state. Each particle stores biological physical attributes drive growth adaptation such as light sensitivity, wind interaction, obstacles. This representation allows external effects can be induced at any time without prior...

We present a novel method for the combustion of botanical tree models. Tree models are represented as connected particles branching structure and polygonal surface mesh combustion. Each particle stores biological physical attributes that drive kinetic behavior plant exothermic reaction Coupled with realistic physics rods, enable dynamic branch motions. model material properties, such moisture charring behavior, associate them individual particles. The is efficiently processed in domain on...

The placement of vegetation plays a central role in the realism virtual scenes. We introduce procedural models (PPMs) for urban layouts. PPMs are environmentally sensitive to city geometry and allow identifying plausible plant positions based on structural functional zones an layout. can either be directly used by defining their parameters or learned from satellite images land register data. This allows us populate landscapes with complex 3D enhance existing approaches generating landscapes....