As deep learning continues to make progress for challenging perception tasks, there is increased interest in combining vision, language, and decision-making. Specifically, the Vision Language Navigation (VLN) task involves navigating a goal purely from language instructions visual information without explicit knowledge of goal. Recent successful approaches have made in-roads achieving good success rates this but rely on beam search, which thoroughly explores large number trajectories...

Although various image-based domain adaptation (DA) techniques have been proposed in recent years, shift videos is still not well-explored. Most previous works only evaluate performance on small-scale datasets which are saturated. Therefore, we first propose two large-scale video DA with much larger discrepancy: UCF-HMDB_full and Kinetics-Gameplay. Second, investigate different integration methods for videos, show that simultaneously aligning learning temporal dynamics achieves effective...

In this paper, we consider the distributed parameter estimation in wireless sensor networks where a total bit rate constraint is imposed. We study optimal tradeoff between number of active sensors and quantization for each to minimize mean-square error (MSE). To facilitate solution, first introduce concept equivalent 1-bit MSE function. Next, present an algorithm homogeneous based on minimizing Then, quasi-optimal heterogeneous networks, which also function, upper bound proposed addressed....

In this paper, we consider distributed estimation of a noise-corrupted deterministic parameter in energy-constrained wireless sensor networks from energy-distortion perspective. Given total energy budget allowable to be used by all sensors, there exists tradeoff between the subset active sensors and each order minimize MSE. To determine optimal quantization bit rate transmission sensor, concept equivalent unit-energy MSE function is introduced. Based on concept, an algorithm for homogeneous...

Human actions often involve complex interactions across several inter-related objects in the scene. However, existing approaches to fine-grained video understanding or visual relationship detection rely on single object representation pairwise relationships. Furthermore, learning multiple hundreds of frames for is computationally infeasible and performance may suffer since a large combinatorial space has be modeled. In this paper, we propose efficiently learn higher-order between arbitrary...

The recent interest in using deep learning for seismic interpretation tasks, such as facies classification, has been facing a significant obstacle, namely, the absence of large publicly available annotated data sets training and testing models. As result, researchers have often resorted to annotating their own data. However, different may annotate classes or use train test splits. In addition, it is common papers that apply machine classification not contain quantitative results, rather rely...

Air-writing refers to writing of linguistic characters or words in a free space by hand finger movements. differs from conventional handwriting; the latter contains pen-up-pen-down motion, while former lacks such delimited sequence events. We address air-writing recognition problems pair companion papers. In Part I, is accomplished based on six-degree-of-freedom motion data. two levels: and words. Isolated can be recognized similar gestures although with increased sophistication variability....

The Vision-and-Language Navigation (VLN) task entails an agent following navigational instruction in photo-realistic unknown environments. This challenging demands that the be aware of which was completed, is needed next, way to go, and its navigation progress towards goal. In this paper, we introduce a self-monitoring with two complementary components: (1) visual-textual co-grounding module locate completed past, required for next action, moving direction from surrounding images (2) monitor...

Despite the recent progress of fully-supervised action segmentation techniques, performance is still not fully satisfactory. One main challenge problem spatiotemporal variations (e.g. different people may perform same activity in various ways). Therefore, we exploit unlabeled videos to address this by reformulating task as a cross-domain with domain discrepancy caused spatio-temporal variations. To reduce discrepancy, propose SelfSupervised Temporal Domain Adaptation (SSTDA), which contains...

PreviousNext No AccessSEG Technical Program Expanded Abstracts 2019Semi-supervised learning for acoustic impedance inversionAuthors: Motaz AlfarrajGhassan AlRegibMotaz AlfarrajCenter Energy and Geo Processing (CeGP), Georgia Institute of TechnologySearch more papers by this author Ghassan AlRegibCenter authorhttps://doi.org/10.1190/segam2019-3215902.1 SectionsSupplemental MaterialAboutPDF/ePub ToolsAdd to favoritesDownload CitationsTrack CitationsPermissions ShareFacebookTwitterLinked...

Three-dimensional television (3DTV) is believed to be the future of broadcasting that would replace current 2D HDTV technology. Future 3DTV bring a more life-like and visually immersive home entertainment experience, in which users will have freedom navigate through scene choose different viewpoint. A desired view can synthesized at receiver side using depth image-based rendering (DIBR). While this approach has many advantages, one key challenges DIBR how fill holes caused by disocclusion...

The identification of salt-dome boundaries in migrated seismic data volumes is important for locating petroleum reservoirs. presence noise the makes computer-aided interpretation even more challenging. We have developed noise-robust algorithms that could label salt domes effectively and efficiently. Our research twofold. First, we used a texture-based gradient to accomplish detection. found by using dissimilarity measure based on 2D discrete Fourier transform, algorithm was capable...

Air-writing refers to writing of characters or words in the free space by hand finger movements. We address air-writing recognition problems two companion papers. Part 2 addresses detecting and recognizing activities that are embedded a continuous motion trajectory without delimitation. Detection intended among superfluous movements unrelated letters presents challenge needs be treated separately from traditional problem pattern recognition. first present dataset contains mixture nonwriting...

Traffic signs are critical for maintaining the safety and efficiency of our roads. Therefore, we need to carefully assess capabilities limitations automated traffic sign detection systems. Existing datasets limited in terms type severity challenging conditions. Metadata corresponding these conditions unavailable it is not possible investigate effect a single factor because simultaneous changes numerous To overcome shortcomings existing datasets, introduced CURE-TSD-Real dataset, which based...

Summary Fault detection is one of the major tasks subsurface interpretation and reservoir characterization from 3D seismic surveying. However, with growing data in both its size resolution, efficiency interpreting faults increasingly relies on development powerful computational tools that are capable mimicking an experienced interpreter's intelligence. In recent years, convolutional neural network (CNN) has been successful for image/video processing various disciplines attracting more...

Reservoir characterization involves the estimation petrophysical properties from well-log data and seismic data. Estimating such is a challenging task due to non-linearity heterogeneity of subsurface. Various attempts have been made estimate using machine learning techniques as feed-forward neural networks support vector regression (SVR). Recent advances in shown promising results for recurrent (RNN) modeling complex sequential videos speech signals. In this work, we propose an algorithm...

Understanding Earth's subsurface structures has been and continues to be an essential component of various applications such as environmental monitoring, carbon sequestration, oil gas exploration. By viewing the seismic volumes that are generated through processing recorded traces, researchers were able learn from applying advanced image computer vision algorithms effectively analyze understand structures. In this article, we first summarize recent advances in direction relied heavily on...

Reliable fault detection is one of the major tasks subsurface interpretation and reservoir characterization from three-dimensional (3D) seismic surveying. This study presents an innovative workflow based on multi-attribute support vector machine (SVM) analysis a volume, which consists four steps. First, three groups attributes are selected computed volume amplitude, including edge-detection, geometric, texture, all clearly highlight faults in attribute images. Second, two sets training...

In exploration seismology, seismic inversion refers to the process of inferring physical properties subsurface from data. Knowledge can prove helpful in identifying key structures for hydrocarbon exploration. this work, we propose a workflow predicting acoustic impedance (AI) data using network architecture based on Temporal Convolutional Network by posing problem as that sequence modeling. The proposed overcomes some problems other architectures usually face, like gradient vanishing...

Fault interpretation is one of the routine processes used for subsurface structure mapping and reservoir characterization from 3D seismic data. Various techniques have been developed computer-aided fault imaging in past few decades; example, conventional methods edge detection, curvature analysis, red-green-blue rendering, popular machine-learning such as support vector machine (SVM), multilayer perceptron (MLP), convolutional neural network (CNN). However, most are performed at sample level...

Cooperative diversity techniques exploit the spatial characteristics of network to create transmit-diversity, in which same information can be forwarded through multiple paths towards a single destination or set nodes. In this paper, we study integration cooperative into wireless routing protocols by developing distributed MAC (C-MAC) and protocols. The proposed employ efficient relay selection-coordination power allocation maximize cooperation benefits network. Simulation results show that...