7903 research outputs found
Sort by
Enhancing VQGAN-Based Model with ConNeXt for Blind Super-Resolution
Get PDFThis thesis presents a novel super-resolution model based on Vector Quantized Generative Adversarial Network (VQGAN) to enhance image resolution. Inspired by recent advancements in the field of image reconstruction, we apply VQGAN to the super-resolution task, leveraging its powerful generative capabilities to produce higher quality high-resolution images.
Building on the VQGAN framework, we propose an improved architecture that incorporates an additional ConvNeXt feature extractor based on Convolutional Neural Networks (CNN) to effectively capture and refine features from low-resolution images. To further enhance model performance, we implemented various strategies to optimize the utilization of the codebook, including capacity optimization, improved initialization, and an Exponential Moving Average (EMA) dynamic updating strategy to ensure more efficient and diverse codebook usage. Additionally, we divided the training process into two phases to improve training stability and convergence. In the first phase, we focus on training the codebook and decoder to create a strong high-resolution prior. In the second phase, we concentrate on training the encoder, the ConvNeXt-based feature extractor, and the GAN. This phased training approach enables the model to progressively learn and refine the complex details required for high-quality image reconstruction and enhances model stability.
Experimental results demonstrate that our model shows improvements to the existing methods in image quality as measured by Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), and Learned Perceptual Image Patch Similarity (LPIPS). This work highlights the potential of combining VQGAN with advanced feature extractors in super-resolution models, paving the way for future research and development in this field
Dolphin: Drone-Based Oceanic Landing Platform with High-Tech Integrated Navigation
Get PDFThe ocean covers 71% of the earth’s surface. With its vast size, it produces almost half of the world’s oxygen, is a source of food, and controls the climate. As humans, it is vital to our way of life. Thus, institutions and organizations have been built in order to study and conserve it; one such organization is the Monterey Bay Aquarium Research Institute (MBARI). A method available to them is the use of drone imaging over the ocean’s surface. However, drones have a limited battery life that restricts the scope of missions they can pursue. A solution that we have provided is DOLPHIN, an autonomous drone landing platform capable of expanding the area that drones are able to study.
The current iteration of DOLPHIN utilizes the vessel of a small-waterplane-area-twin-hull (SWATH) design of a previous senior design project started in 2005 and builds upon it. Repairs to its frame were made to ensure it is seaworthy. DOLPHIN consists of three main subsystems: propulsion, automated navigation and the landing platform. All three subsystems were developed into a cohesive design. The propulsion subsystem focused on the motor controller used, thrusters used, thruster layout, as well as the number of thrusters used. The second subsystem, the automated navigation system, focused on developing capabilities such as Global Positioning System (GPS) waypoint guidance, holding position capabilities, and a switch between manual control and dead modes. The third subsystem, the landing platform, focused on developing the platform itself, which is stabilized using linear actuators, and the frame that supported the assembly. Frequent tests at Alviso Boat Ramp and in Bethany Reservoir confirmed an acceptable stabilizing landing platform and autopilot with station keeping. Further developments to the project would include improving station keeping capabilities and refining the subsystem’s electrical and software features
Reflections on Migrant Farm Working Families
Get PDFWhen I think about Silicon Valley and the big tech companies that are only about a thirty-minute drive from Half Moon Bay, there are major differences between the pay, benefits, and perks that come with working at the goggles of the world. Thermos for work. Boots for the rain. Rain Gear. Hoodies to protect us from the sun and to keep us warm are some of the things that the campesinos asked for… I think of corporate jobs and how when you get hired you get all the material that you will need to do your job given to you. Campesinos on the other hand make very little money and have to still buy all of their material that would allow them to do the job. This is something that I ask myself to this day. Why do the people who get paid the most get access to free transportation, free food, a free laptop, a chair for your home office, and a monitor to better see your work yet the people who make the least have to use from the little that they make to buy all of these things. I hope to continue to have these informal and informal conversations with the campesinos so that I can record as many of these stories as possible and also document the need to repost back as soon as possible to our audience of supporters. I hope to one day also be able to use…my future research to put together a proposal for our local, state, and national political leaders. I strongly believe that this should be a situation that should be taken care of from the root and that is passing legislation to better treat our Farmworkers
Legislating Morality in the Gilded Age and Progressive Era: Moral Panic and the “White Slave” Case That Changed America
Get PDFThis article is based on the presidential address presented to the Society for Historians of the Gilded Age and Progressive Era at the meeting of the Organization of American Historians in Los Angeles in 2023. Its focus is Maury Diggs and Drew Caminetti, two white men from Sacramento, California, charged with violating the Mann Act (known as the White Slave Trafficking Act) in 1913. The Gilded Age and Progressive Era obsession with white slavery, a phenomenon that has particular resonance in today’s climate, reveals the power of moral panics. Examining the steps, and missteps, that various legal, social, and political entities, including all three branches of government, took in response to Diggs and Caminetti’s actions highlights some of the major social changes gripping the nation. Moral panics can be investigated as crucial historical sites of contestation, revealing efforts to neutralize or turn back the societal changes perceived to be the greatest threat to the prevailing social power structure—in this case foreigners, the new leisure culture, the liberalization of sexual attitudes, and the threat of female independence. Understanding the origins and repercussions of past moral panics can help identify, understand, and possibly defuse future panics
Sheep Camp Creek Low-water Crossing Redesign
Get PDFThe San Francisco Public Utilities Commission (SFPUC) owns a biohabitat restoration site in Sunol, California titled the Sheep Camp Creek Biohabitat Restoration Site. The goal of this site is to provide habitat to endangered species and improve the environmental benefits of the area. The site is split in half by the main basin stream, which was sectioned-off as part of the restoration efforts. Three cattle crossings are still present on-site, however, to facilitate travel between either side of the site. The most downstream crossing present on-site, however, has seen detrimental erosive effects as a result of major storms that have caused the need for replacement three times in the past decade. This project conducted a hydrologic and hydraulic analysis of the site to determine the best crossing solution for the site at this location. The project included a point-precipitation model to provide basin-wide hydrologic information to the SFPUC, as well as existing and proposed conditions hydraulic models developed in HEC-RAS to view hydraulic effects of different crossing solutions. Finally, cost estimates and life cycle analyses were prepared for each solution. These results were then presented to the SFPUC for their review to aid future planning decisions
Deep Learning Based Single Image Super-resolution
Get PDFSingle image super-resolution (SR) involves taking a given low-resolution (LR) image and generating a corresponding high-resolution (HR) image. This is a core task in the computer vision field due to its multitude of applications ranging from helping current-day issues of storage and transfer of data to restoration of low-resolution images. The current strategies, however, struggle to reach the quality needed for their widespread use and are often too resource-intensive for the average consumer. While other lightweight SR techniques exist with different techniques, like Cascading Residual Networks [3], their success comes at the cost of expensive technology inaccessible to most situations. With this in mind, we propose a model built to be a lightweight SR network while still improving the quality of the image. This was achieved by using an existing lightweight SR network, SwinIR [9], as the generator. To increase the quality, we used a Generative Adversarial Network (GAN) based on SR360 [3] with a lightweight discriminator, MobileViT [11] to improve upon the quality without sacrificing the lightweight nature of SwinIR. This resulted in a generator that is lightweight and gives high-quality results that was submitted to the NTIRE 2024 Image Super-Resolution (x4)
Nautilus: Deep Sea Dexterity
Get PDFGlobal warming and climate change are prevalent issues in today\u27s society. As a result, research in the ocean, earth\u27s biggest ecosystem, is imperative in efforts to protect the environment. Santa Clara University\u27s Robotic Systems Lab contributes to this field through work and development on remotely operated vehicles (ROVs). Nautilus was built to support geologists and other researchers, primarily in the Lake Tahoe and Monterey Bay area. Thus far the ROV s sole scientific instrument has been its onboard camera, though partner geologists desire the ability to collect physical samples using a robotic arm attachment at depths up to 100 meters. In direct collaboration with researchers within the Monterey Bay Aquarium Research Institute (MBARI), the team was able to create and accomplish a set of deliverables to improve the ROV. The team\u27s goal was to make the overall experience of the pilot a more user-friendly experience. This was done by creating an entirely new graphical user interface (GUI), to give pilots a better experience when piloting the craft. It also involved an overhaul of last year\u27s robotic arm, by mounting the arm on a modular tool sled as well as upgrading all the arm materials to make the system extremely durable. Unfortunately, while the creation of the tool sled was successful, more work is required to develop complete end-point control. Furthermore, designs to improve upon the design of the end-effector were unsuccessful and require more work as well. Finally, various control systems have been added that give depth, heading, altitude, and, in the future, end-point control. These features take much of the strain of operating an ROV off of the pilot, allowing them to focus on collecting data by locking certain parameters that they would otherwise have to be conscious of when piloting the ROV. The project has resulted in a more robust and easier to use ROV that can continue to support exploration and documentation of lakes and shallow stretches of the ocean far more easily
SCDI: Privacy-Preserving Device Identification
No full textAs internet security concerns grow due to the increasing complexity of cyberattacks, heterogeneity of network-connected devices, and popularity of insecure Internet of Things (IoT) devices, network administrators require more intelligent ways of managing and securing their networks. One crucial step in this regard is the ability to identify and classify devices from the edge of networks. However, many recent advances in network security come at the cost of user privacy, as device classification algorithms often require packet payload data, Internet Protocol (IP) addresses, or other sensitive information. This paper proposes the Shenassa-Castillo Device Identifier (SCDI): a software network device identifier that does not use any potentially sensitive features in classification. The novel feature used for device classification is Acknowledgment Response Time (ART)—that is, the time between a Transmission Control Protocol (TCP) packet and its related acknowledgment. We examine the efficacy ART in distinguishing between heterogeneous devices and its shortcomings in distinguishing between homogeneous devices. We train a Random Forest classifier on a dataset of seven devices and achieve accuracy and F1 scores of 88.3% and 87.2%, respectively
