170,051 research outputs found
Changes in the expression of homeobox transcription factors OTX1 and OTX2 in the rat myenteric plexus after DNBS-induced colitis
No full textA robust feature tracker for active surveillance of outdoor scenes
No full textIn this paper, we propose a robust real-time object detection system for outdoor image sequences acquired by an active camera. The system is able to compensate background changes due to the camera motion and to detect mobile objects in the scene. Background compensation is performed by assuming a simple translation (displacement vector) of the background from the previous to the current frame and by applying the well-known tracker proposed by Lucas and Kanade. A reference map containing all well trackable features is maintained and updated by the system at each frame by introducing new good features related to new regions that appear in the current image. A new method is applied to reject badly tracked features. The current frame and the background after compensation are processed by a change detection method in order to locate mobile objects. Results are presented in the contest of a visual-based surveillance system for monitoring outdoor enviroments
Memory Driven Design Methodologies for Optimal SSD Performance
No full textSolid State Drives (SSDs) are one of the electronic systems with the higher development rate in the last decade: they are widely used in hyper scale systems such as cloud computing and big data servers where performance is a constraint, as well as in consumer electronics by replacing traditional hard disk drives (HDDs)
Self Attention based multi branch Network for Person Re-Identification
No full textRecent progress in the field of person re-identification have shown promising improvement by designing neural networks to learn most discriminative features representations. Some efforts utilize similar parts from different locations to learn better representation with the help of soft attention, while others search for part based learning methods to enhance consecutive regions relationships in the learned features. However, only few attempts have been made to learn non-local similar parts directly for the person re-identification problem. In this paper, we propose a novel self attention based multi branch(classifier) network to directly model long-range dependencies in the learned features. Multi classifiers assist the model to learn discriminative features while self attention module encourages the learning to be independent of the feature map locations. Spectral normalization is applied in the whole network to improve the training dynamics and for the better convergence of the model. Experimental results on two benchmark datasets have shown the robustness of the proposed work
Visual tracking in camera-switching outdoor sport videos: Benchmark and baselines for skiing
No full textSkiing is a globally popular winter sport discipline with a rich history of competitive events. This domain offers ample opportunities for the application of computer vision to enhance the understanding of athletes’ performances. However, this potential has remained relatively untapped in comparison to other sports, primarily due to the limited availability of dedicated research studies and datasets. The present paper takes a significant stride towards bridging these gaps. It conducts a comprehensive examination of skier appearance tracking in videos capturing their entire performance—an essential step for more advanced performance analyses. To implement this investigation, we introduce SkiTB, the largest and most annotated dataset tailored for computer vision applications in skiing. We subject a range of visual object tracking algorithms to rigorous testing, including both well-established methodologies and a novel skier-specific baseline algorithm. The results yield valuable insights into the suitability of various tracking techniques for vision-based skiing analysis and into the generalization of state-of-the-art algorithms to complex target behaviors and conditions set by winter outdoor environments. To foster further development, we make SkiTB, the associated code, and the obtained results accessible through https://machinelearning.uniud.it/datasets/skitb
Machine Learning and Non-volatile Memories
No full textThis book presents the basics of both NAND flash storage and machine learning, detailing the storage problems the latter can help to solve. At a first sight, machine learning and non-volatile memories seem very far away from each other. Machine learning implies mathematics, algorithms and a lot of computation; non-volatile memories are solid-state devices used to store information, having the amazing capability of retaining the information even without power supply. This book will help the reader understand how these two worlds can work together, bringing a lot of value to each other. In particular, the book covers two main fields of application: analog neural networks (NNs) and solid-state drives (SSDs).
After reviewing the basics of machine learning in Chapter 1, Chapter 2 shows how neural networks can mimic the human brain; to accomplish this result, neural networks have to perform a specific computation called vector-by-matrix (VbM) multiplication, which is particularly power hungry. In the digital domain, VbM is implemented by means of logic gates which dictate both the area occupation and the power consumption; the combination of the two poses serious challenges to the hardware scalability, thus limiting the size of the neural network itself, especially in terms of the number of processable inputs and outputs. Non-volatile memories (phase change memories in Chapter 3, resistive memories in Chapter 4, and 3D flash memories in Chapter 5 and Chapter 6) enable the analog implementation of the VbM (also called “neuromorphic architecture”), which can easily beat the equivalent digital implementation in terms of both speed and energy consumption.
SSDs and flash memories are strictly coupled together; as 3D flash scales, there is a significant amount of work that has to be done in order to optimize the overall performances of SSDs. Machine learning has emerged as a viable solution in many stages of this process. After introducing the main flash reliability issues, Chapter 7 shows both supervised and un-supervised machine learning techniques that can be applied to NAND. In addition, Chapter 7 deals with algorithms and techniques for a pro-active reliability management of SSDs. Last but not least, the last section of Chapter 7 discusses the next challenge for machine learning in the context of the so-called computational storage.
No doubt that machine learning and non-volatile memories can help each other, but we are just at the beginning of the journey; this book helps researchers understand the basics of each field by providing real application examples, hopefully, providing a good starting point for the next level of development
Improving 3D NAND Flash Memories Reliability: a Cross-Layer Perspective
No full textThe 3D NAND Flash memory technology is the main building block of storage architectures such as multimedia cards and Solid-State Drives. Applications ranging from mobile to high-performance computing are continuously calling for an increased storage density, requiring massive scaling efforts at the device and array level. This leads to a natural degradation of the functional metrics of the technology while exposing new reliability issues that jeopardize the inherent memory trade-off with performance. A simple device optimization would be insufficient to tackle the problem. In this work, we show through some case studies how a cross-layer approach spanning from devices and circuits to system-level optimizations is mandatory for future storage systems development
Editorial for the special issue on flash memory devices
No full textFlash memory devices represented a breakthrough in the storage industry since their inception in the mid-1980s, and innovation is still ongoing after more than 35 years. This Special Issue provides insight on and advancements in Flash memory devices
Real image super-resolution using GAN through modeling of LR and HR process.
No full textThe current existing deep image super-resolution methods usually assume that a Low Resolution (LR) image is bicubicly downscaled of a High Resolution (HR) image. However, such an ideal bicubic downsampling process is different from the real LR degradations, which usually come from complicated combinations of different degradation processes, such as camera blur, sensor noise, sharpening artifacts, JPEG compression, and further image editing, and several times image transmission over the internet and unpredictable noises. It leads to the highly ill-posed nature of the inverse upscaling problem. To address these issues, we propose a GAN-based SR approach with learnable adaptive sinusoidal nonlinearities incorporated in LR and SR models by directly learn degradation distributions and then synthesize paired LR/HR training data to train the generalized SR model to real image degradations. We demonstrate the effectiveness of our proposed approach in quantitative and qualitative experiments
- …
