Sequence-to-sequence vision-language models are showing promise, but their applicability is limited by their inference latency due to their autoregressive way of generating predictions. We propose a parallel decoding sequence-to-sequence vision-language model, trained with a Query-CTC loss, that marginalizes over multiple inference paths in the decoder. This allows us to model the joint distribution of tokens, rather than restricting to conditional distribution as in an autoregressive model. The resulting model, NARVL, achieves performance on-par with its state-of-the-art autoregressive counterpart, but is faster at inference time, reducing from the linear complexity associated with the sequential generation of tokens to a paradigm of constant time joint inference.

Increasingly mobile data traffic and high quality service demand has driven standard developments and new mobile technologies deployment at an unprecedented level. Long Term Evolution (LTE) standard and its improved version LTE-Advanced (LTE-A) are two technology standards developed to cope with high levels of mobile data traffic demand. However, traffic and revenue disparity still is a reality, suggesting that traditional network deployment methods - based mainly on macro cellular sites - might prove to be cost ineffective in the long term. From another perspective, and increasingly important for mobile network operators, revenue is also a function of each mobile network deployment's sustainability. This work aims to comprehensively elaborate on those matters by presenting four specific scenarios with a comparative analysis of both macro and femtocell deployments (single and both technology networks). For each scenario, capacity, cost effectiveness and expected carbon emissions are the evaluated key indicators. This kind of analysis provides mobile networks operators with relevant information, enabling them to sustainably adapt business and provisioning models as well as network deployment strategies to current and future technological standards, while minimizing capital and operational expenditure (CAPEX/OPEX). The main contribution is that in short term, mixed macro and femtocell deployment scenarios are the most cost effective and sustainable option, while in mid to long term, as data traffic demand rises, femtocell deployments become the most sustainable, not only from economical and environmental points of view, but also from network coverage stand point.

This paper presents an extension to David Lowe’s well-known object recognition algorithm based on his Scale and Feature Invariant Transform (SIFT). One of the benefits of Lowe’s SIFT-based method is that it can recognize objects from only three keypoints. While this capability can be useful in circumstances where the cost of a false negative is high, it’s often the case that false positive are an equal, or greater, concern. We extend Lowe’s algorithm by adding the ability to use 3D constraints during matching. These constraints essentially eliminate false positive matches. We combine our extension with the original algorithm to retain the recognition power of the original method while adding significant robustness against false positives, thereby increasing overall classification power. In addition to improving recognition, our extension returns 3D pose information. Yet, it adds very little computational overhead to Lowe’s original algorithm. 1

We describe a method of generating and utilizing visual landmarks that is well suited for SLAM applications. The landmarks created are highly distinctive and reliably detected, virtually eliminating the data association problem present in other landmark schemes. Upon subsequent detections of a landmark, a 3-D pose can be estimated. The scheme requires a single camera.

This paper presents the Visual Simultaneous Localization and Mapping (vSLAMTM) algorithm, a novel algorithm for simultaneous localization and mapping (SLAM). The algorithm is vision-and odometry-based, and enables low-cost navigation in cluttered and populated environments. No initial map is required, and it satisfactorily handles dynamic changes in the environment, for example, lighting changes, moving objects and/or people. Typically, vSLAM recovers quickly from dramatic disturbances, such as “kidnapping”.

This paper presents a system for automatic reconstruction of 3D structure using two unsynchronized cameras. Three images are acquired sequentially from the left, right, and again from the left camera. A virtual image from the left camera synchronized with the right image is created by interpolating matching points of interest (SIFT features) in the two left images. Both geometric and probabilistic criteria are used to select the correct set of matching features amongst the three views. In an indoor environment, the method typically results in 3D structure with approximately 200 feature points, with a median 3D accuracy of 1.6 cm when the average depth is 3 m and the robot has moved 1-2 cm between each image acquisition.

Service robotics products are becoming a reality. This paper describes three core technologies that enable the next generation of service robots. They are low-cost, make use of low-cost hardware, and prepare for a short time-to-market for product development. The first technology is an object recognition system, which can be used by the robot to interact with the environment The second technology is a vision-based navigation system (vSLAM/spl trade/), which simultaneously can build a map and localize the robot in the map. Finally, the third technology is a flexible and rich software platform (ERSP/spl trade/) that assists developers in rapid design and prototyping of robotics applications.

An unsupervised learning algorithm that can obtain a probabilistic model of an object composed of a collection of parts (a moving human body in our examples) automatically from unlabeled training data is presented. The training data include both useful "foreground" features as well as features that arise from irrelevant background clutter - the correspondence between parts and detected features is unknown. The joint probability density function of the parts is represented by a mixture of decomposable triangulated graphs which allow for fast detection. To learn the model structure as well as model parameters, an EM-like algorithm is developed where the labeling of the data (part assignments) is treated as hidden variables. The unsupervised learning technique is not limited to decomposable triangulated graphs. The efficiency and effectiveness of our algorithm is demonstrated by applying it to generate models of human motion automatically from unlabeled image sequences, and testing the learned models on a variety of sequences.

Decomposable triangulated graphs have been shown to be efficient and effective for modeling the probabilistic spatio-temporal structure of brief stretches of human motion. In previous work such model structure was handcrafted by expert human observers and labeled data were needed for parameter learning. We present a method to build automatically the structure of the decomposable triangulated graph from unlabeled data. It is based on maximum-likelihood. Taking the labeling of the data as hidden variables, a variant of the EM algorithm can be applied. A greedy algorithm is developed to search for the optimal structure of the decomposable model based on the (conditional) differential entropy of variables. Our algorithm is demonstrated by learning models of human motion completely automatically from unlabeled real image sequences with clutter and occlusion. Experiments on both motion captured data and grayscale image sequences show that the resulting models perform better than the hand-constructed models.

This paper presents an unsupervised learning algorithm that can derive the probabilistic dependence structure of parts of an object (a moving hu- man body in our examples) automatically from unlabeled data. The dis- tinguished part of this work is that it is based on unlabeled data, i.e., the training features include both useful foreground parts and background clutter and the correspondence between the parts and detected features are unknown. We use decomposable triangulated graphs to depict the probabilistic independence of parts, but the unsupervised technique is not limited to this type of graph. In the new approach, labeling of the data (part assignments) is taken as hidden variables and the EM algo- rithm is applied. A greedy algorithm is developed to select parts and to search for the optimal structure based on the differential entropy of these variables. The success of our algorithm is demonstrated by applying it to generate models of human motion automatically from unlabeled real image sequences.

Computer perception of biological motion is key to developing convenient and powerful human-computer interfaces. Successful body tracking algorithms have been developed; however initialization is done by hand. We propose a method for detecting a moving human body and for labeling its parts automatically. It is based on maximizing the joint probability density function (PDF) of the position and velocity of the body parts. The PDF is estimated from training data. Dynamic programming is used for calculating efficiently the best global labeling on an approximation of the PDF. The computational cost is on the order of N/sup 4/ where N is the number of features detected. We explore the performance of our method with experiments carried on a variety of periodic and non-periodic body motions viewed monocularly for a total of approximately 30,000 frames. Point-markers were strapped to the joints of the subject for facilitating image analysis. We find an average of 2.3% labeling error; the experiments also suggest a high degree of viewpoint-invariance.

We have developed a system capable of tracking a human arm in 3D and in real time. The system is based on a previously developed algorithm for 3D tracking which requires only a monocular view and no special markers on the body. In this paper we describe our real-time system and the insights gained from real-time experimentation.

We address the problem of estimating the position and motion of a human arm in 3D without any constraints on its behavior and without the use of special markers. We model the arm as two truncated right-circular cones connected with spherical joints. We propose to use a recursive estimator for arm position, and to provide the estimator with error signals obtained by comparing the projected estimated arm position with that of the actual arm in the image. The system is demonstrated and tested on a real image sequence.>