Pixel-level labelling tasks, such as semantic segmentation, play a central role in image understanding. Recent approaches have attempted to harness the capabilities of deep learning techniques for image recognition to tackle pixellevel labelling tasks. One central issue in this methodology is the limited capacity of deep learning techniques to delineate visual objects. To solve this problem, we introduce a new form of convolutional neural network that combines the strengths of Convolutional Neural Networks (CNNs) and Conditional Random Fields (CRFs)-based probabilistic graphical modelling. To this end, we formulate mean-field approximate inference for the Conditional Random Fields with Gaussian pairwise potentials as Recurrent Neural Networks. This network, called CRF-RNN, is then plugged in as a part of a CNN to obtain a deep network that has desirable properties of both CNNs and CRFs. Importantly, our system fully integrates CRF modelling with CNNs, making it possible to train the whole deep network end-to-end with the usual back-propagation algorithm, avoiding offline post-processing methods for object delineation.We apply the proposed method to the problem of semantic image segmentation, obtaining top results on the challenging Pascal VOC 2012 segmentation benchmark.

No abstract

Detecting visually salient regions in images is

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

a b s t r a c tPockmarks form where fluids discharge through seafloor sediments rapidly enough to make them quick, and are common where gas is present in near-seafloor sediments. This paper investigates how gas might lead to pockmark formation. The process is envisioned as follows: a capillary seal traps gas beneath a fine-grained sediment layer or layers, perhaps layers whose pores have been reduced in size by hydrate crystallization. Gas accumulates until its pressure is sufficient for gas to invade the seal. The seal then fails completely (a unique aspect of capillary seals), releasing a large fraction of the accumulated gas into an upward-propagating gas chimney, which displaces water like a piston as it rises. Near the seafloor the water flow causes the sediments to become ''quick'' (i.e., liquefied) in the sense that grain-to-grain contact is lost and the grains are suspended dynamically by the upward flow. The quickened sediment is removed by ocean-bottom currents, and a pockmark is formed. Equations that approximately describe this gas-piston-water-drive show that deformation of the sediments above the chimney and water flow fast enough to quicken the sediments begins when the gas chimney reaches half way from the base of its source gas pocket to the seafloor. For uniform near-surface sediment permeability, this is a buoyancy control, not a permeability control. The rate the gas chimney grows depends on sediment permeability and the ratio of the depth below seafloor of the top of the gas pocket to the thickness of the gas pocket at the time of seal failure. Plausible estimates of these parameters suggest gas chimneys at Blake Ridge could reach the seafloor in less than a decade or more than a century, depending mainly on the permeability of the deforming near-surface sediments. Since these become quick before gas is expelled, gas venting will not provide a useful warning of the seafloor instabilities that are related to pockmark formation. However, detecting gas chimney growth might be a useful risk predictor. Any area underlain by a gas chimney that extends half way or more to the surface should be avoided.

Dense pixel-wise image prediction has been advanced by harnessing the capabilities of Fully Convolutional Networks (FCNs). One central issue of FCNs is the limited capacity to handle joint upsampling. To address the problem, we present a novel building block for FCNs, namely guided filtering layer, which is designed for efficiently generating a highresolution output given the corresponding low-resolution one and a high-resolution guidance map. Such a layer contains learnable parameters, which can be integrated with FCNs and jointly optimized through end-to-end training. To further take advantage of end-to-end training, we plug in a trainable transformation function for generating the task-specific guidance map. Based on the proposed layer, we present a general framework for pixel-wise image prediction, named deep guided filtering network (DGF). The proposed network is evaluated on five image processing tasks. Experiments on MIT-Adobe FiveK Dataset demonstrate that DGF runs 10-100 times faster and achieves the state-of-theart performance. We also show that DGF helps to improve the performance of multiple computer vision tasks.

No abstract

Recently, smart cities, smart homes, and smart medical systems have challenged the functionality and connectivity of the large-scale Internet of Things (IoT) devices. Thus, with the idea of offloading intensive computing tasks from them to edge nodes (ENs), edge computing emerged to supplement these limited devices. Benefit from this advantage, IoT devices can save more energy and still maintain the quality of the services they should provide. However, computational offload decisions involve federation and complex resource management and should be determined in the real-time face to dynamic workloads and radio environments. Therefore, in this work, we use multiple deep reinforcement learning (DRL) agents deployed on multiple edge nodes to indicate the decisions of the IoT devices. On the other hand, with the aim of making DRL-based decisions feasible and further reducing the transmission costs between the IoT devices and edge nodes, federated learning (FL) is used to train DRL agents in a distributed fashion. The experimental results demonstrate the effectiveness of the decision scheme and federated learning in the dynamic IoT system.INDEX TERMS Federated learning, computation offloading, IoT, edge computing.