Not all semantics become confusing when deploying a semantic segmentation model for real-world scene understanding of adverse weather. The true semantics of most pixels have a high likelihood of appearing in the few top classes according to confidence ranking. In this paper, we replace the one-hot pseudo label with a candidate label set (CLS) that consists of only a few ambiguous classes and exploit its effects on self-training-based unsupervised domain adaptation. Specifically, we formulate the problem as a coarse-to-fine process. In the coarse-level process, adaptive CLS selection is proposed to pick a minimal set of confusing candidate labels based on the reliability of label predictions. Then, representation learning and label rectification are iteratively performed to facilitate feature clustering in an embedding space and to disambiguate the confusing semantics. Experimentally, our method outperforms the state-of-the-art methods on three realistic foggy benchmarks.

Abstract Text detection methods based on grouping characters have emerged and have achieved promising performance. Nevertheless, previous methods that grouped characters by learning the relation of adjacent characters or used the heuristic clustering method with a handcrafted feature are unsuitable for dense, curved, or long texts. An effective manner of grouping characters is proposed by introducing K‐Means that is modified by the law of universal gravitation, an outlier detection mechanism and sufficient context information. Based on that, corresponding text detector is presented, named the Text Detector, based on modified K‐Means (KText), which can generate the bounding boundary of word‐level texts with an arbitrary shape. In the experimental stage, two novel stratagems are presented to replenish character‐level annotations to several datasets that provide only word‐level annotations. To evaluate the effectiveness of the method, experiments are carried out on three benchmarks, ICDAR2013, ICDAR2015 and Total‐Text, which contain horizontal, oriented and curved text. The results show that KText performs more competently than most state‐of‐the‐art text detectors when handling dense texts with an arbitrary shape.

Understanding foggy image sequence in driving scene is critical for autonomous driving, but it remains a challenging task due to the difficulty in collecting and annotating real-world images of adverse weather. Recently, self-training strategy has been considered as a powerful solution for unsupervised domain adaptation, which iteratively adapts the model from the source domain to the target domain by generating target pseudo labels and re-training the model. However, the selection of confident pseudo labels inevitably suffers from the conflict between sparsity and accuracy, both of which will lead to suboptimal models. To tackle this problem, we exploit the characteristics of the foggy image sequence of driving scenes to densify the confident pseudo labels. Specifically, based on the two discoveries of local spatial similarity and adjacent temporal correspondence of the sequential image data, we propose a novel Target-Domain driven pseudo label Diffusion (TDo-Dif) scheme. It employs superpixels and optical flows to identify the spatial similarity and temporal correspondence, respectively, and then diffuses the confident but sparse pseudo labels within a superpixel or a temporal corresponding pair linked by the flow. Moreover, to ensure the feature similarity of the diffused pixels, we introduce local spatial similarity loss and temporal contrastive loss in the model re-training stage. Experimental results show that our TDo-Dif scheme helps the adaptive model achieve 51.92% and 53.84% mean intersection-over-union (mIoU) on two publicly available natural foggy datasets (Foggy Zurich and Foggy Driving), which exceeds the state-of-the-art unsupervised domain adaptive semantic segmentation methods. The proposed method can also be applied to non-sequential images in the target domain by considering only spatial similarity.

Office machines (such as printers, scanners, facsimile machines, and copiers) can consume significant amounts of power. Most office machines have sleep modes to save power. Power management of these machines is usually timeout-based: a machine sleeps after being idle long enough. Setting the time-out duration can be difficult: if it is too long, the machine wastes power during idleness. If it is too short, the machine sleeps too soon and too often—the wake-up delay can significantly degrade productivity. Thus, power management is a tradeoff between saving energy and keeping response time short. Many power management policies have been published and one policy may outperform another in some scenarios. There is no definite conclusion regarding which policy is always better. This article describes two methods for office equipment power management. The first method adaptively reduces power based on a constraint of the wake-up delay. The second is a hybrid method with multiple candidate policies and it selects the most appropriate power management policy. Using 6 months of request traces from 18 different printers, we demonstrate that the hybrid policy outperforms individual policies. We also discover that power management based on business hours does not produce consistent energy savings.

The visual data generated by network cameras can be valuable for a wide range of scientific studies such as weather, wildlife, and traffic. The resource demands for analysis of the data may fluctuate significantly for some of these studies (for example, seasonal or during only rush hours). Cloud computing's pay-per-use can be a preferred solution for analysing large amounts of data from these network cameras. There are few studies focusing on how to allocate cloud resources to analyse many video streams from network cameras. Existing autoscaling and load balancing methods are inapplicable due to the nature of the workloads. This paper presents a solution to manage cloud resources analysing multiple video streams. When an analysis program is launched, the resource manager adjusts the resource allocation by one of two methods: gradually scaling up the number of instances or predicting the needed number of instances. Then, the resource manager monitors the utilization and scales the resources in response to the analysis program's loads. The proposed resource manager has been implemented using Microsoft Azure and demonstrated that the system can adaptively adjust resource to cater to the workload. We explore the trade-offs inherent to both methods. Gradually scaling up can reach more efficient resource allocation, but may take multiple trials. The prediction method is faster in making allocation decisions, but may need additional adjustment. We also investigate the impact of the types of virtual machines and find that smaller sizes are usually better in terms of performance-cost ratios.