Vision transformer

A vision transformer (ViT) is a transformer designed for computer vision. A ViT decomposes an input image into a series of patches (rather than text into tokens), serializes each patch into a vector, and maps it to a smaller dimension with a single matrix multiplication. These vector embeddings are then processed by a transformer encoder as if they were token embeddings. ViTs were designed as alternatives to convolutional neural networks (CNNs) in computer vision applications. They have different inductive biases, training stability, and data efficiency. Compared to CNNs, ViTs are less data efficient, but have higher capacity. Some of the largest modern computer vision models are ViTs, such as one with 22B parameters. Subsequent to its publication, many variants were proposed, with hybrid architectures with both features of ViTs and CNNs. ViTs have found application in image recognition, image segmentation, weather prediction, and autonomous driving. == History == Transformers were introduced in Attention Is All You Need (2017), and have found widespread use in natural language processing. A 2019 paper applied ideas from the Transformer to computer vision. Specifically, they started with a ResNet, a standard convolutional neural network used for computer vision, and replaced all convolutional kernels by the self-attention mechanism found in a Transformer. It resulted in superior performance. However, it is not a Vision Transformer. In 2020, an encoder-only Transformer was adapted for computer vision, yielding the ViT, which reached state of the art in image classification, overcoming the previous dominance of CNN. The masked autoencoder (2022) extended ViT to work with unsupervised training. The vision transformer and the masked autoencoder, in turn, stimulated new developments in convolutional neural networks. Subsequently, there was cross-fertilization between the previous CNN approach and the ViT approach. In 2021, some important variants of the Vision Transformers were proposed. These variants are mainly intended to be more efficient, more accurate or better suited to a specific domain. Two studies improved efficiency and robustness of ViT by adding a CNN as a preprocessor. The Swin Transformer achieved state-of-the-art results on some object detection datasets such as COCO, by using convolution-like sliding windows of attention mechanism, and the pyramid process in classical computer vision. == Overview == The basic architecture, used by the original 2020 paper, is as follows. In summary, it is a BERT-like encoder-only Transformer. The input image is of type R H × W × C {\displaystyle \mathbb {R} ^{H\times W\times C}} , where H , W , C {\displaystyle H,W,C} are height, width, channel (RGB). It is then split into square-shaped patches of type R P × P × C {\displaystyle \mathbb {R} ^{P\times P\times C}} . For each patch, the patch is pushed through a linear operator, to obtain a vector ("patch embedding"). The position of the patch is also transformed into a vector by "position encoding" (the paper tried no embedding, 1D embedding, 2D embedding, and relative embedding: 1D was adopted). The two vectors are added, then pushed through several Transformer encoders. The attention mechanism in a ViT repeatedly transforms representation vectors of image patches, incorporating more and more semantic relations between image patches in an image. This is analogous to how in natural language processing, as representation vectors flow through a transformer, they incorporate more and more semantic relations between words, from syntax to semantics. The above architecture turns an image into a sequence of vector representations. To use these for downstream applications, an additional head needs to be trained to interpret them. For example, to use it for classification, one can add a shallow MLP on top of it that outputs a probability distribution over classes. The original paper uses a linear-GeLU-linear-softmax network. == Variants == === Original ViT === The original ViT was an encoder-only Transformer supervise-trained to predict the image label from the patches of the image. As in the case of BERT, it uses a special token in the input side, and the corresponding output vector is used as the only input of the final output MLP head. The special token is an architectural hack to allow the model to compress all information relevant for predicting the image label into one vector. Transformers found their initial applications in natural language processing tasks, as demonstrated by language models such as BERT and GPT-3. By contrast the typical image processing system uses a convolutional neural network (CNN). Well-known projects include Xception, ResNet, EfficientNet, DenseNet, and Inception. Transformers measure the relationships between pairs of input tokens (words in the case of text strings), termed attention. The cost is quadratic in the number of tokens. For images, the basic unit of analysis is the pixel. However, computing relationships for every pixel pair in a typical image is prohibitive in terms of memory and computation. Instead, ViT computes relationships among pixels in various small sections of the image (e.g., 16x16 pixels), at a drastically reduced cost. The sections (with positional embeddings) are placed in a sequence. The embeddings are learnable vectors. Each section is arranged into a linear sequence and multiplied by the embedding matrix. The result, with the position embedding is fed to the transformer. === Architectural improvements === ==== Pooling ==== After the ViT processes an image, it produces some embedding vectors. These must be converted to a single class probability prediction by some kind of network. In the original ViT and Masked Autoencoder, they used a dummy [CLS] token, in emulation of the BERT language model. The output at [CLS] is the classification token, which is then processed by a LayerNorm-feedforward-softmax module into a probability distribution. Global average pooling (GAP) does not use the dummy token, but simply takes the average of all output tokens as the classification token. It was mentioned in the original ViT as being equally good. Multihead attention pooling (MAP) applies a multiheaded attention block to pooling. Specifically, it takes as input a list of vectors x 1 , x 2 , … , x n {\displaystyle x_{1},x_{2},\dots ,x_{n}} , which might be thought of as the output vectors of a layer of a ViT. The output from MAP is M u l t i h e a d e d A t t e n t i o n ( Q , V , V ) {\displaystyle \mathrm {MultiheadedAttention} (Q,V,V)} , where q {\displaystyle q} is a trainable query vector, and V {\displaystyle V} is the matrix with rows being x 1 , x 2 , … , x n {\displaystyle x_{1},x_{2},\dots ,x_{n}} . This was first proposed in the Set Transformer architecture. Later papers demonstrated that GAP and MAP both perform better than BERT-like pooling. A variant of MAP was proposed as class attention, which applies MAP, then feedforward, then MAP again. Re-attention was proposed to allow training deep ViT. It changes the multiheaded attention module. === Masked Autoencoder === The Masked Autoencoder took inspiration from denoising autoencoders and context encoders. It has two ViTs put end-to-end. The first one ("encoder") takes in image patches with positional encoding, and outputs vectors representing each patch. The second one (called "decoder", even though it is still an encoder-only Transformer) takes in vectors with positional encoding and outputs image patches again. ==== Training ==== During training, input images (224px x 224 px in the original implementation) are split along a designated number of lines on each axis, producing image patches. A certain percentage of patches are selected to be masked out by mask tokens, while all others are retained in the image. The network is tasked with reconstructing the image from the remaining unmasked patches. Mask tokens in the original implementation are learnable vector quantities. A linear projection with positional embeddings is then applied to the vector of unmasked patches. Experiments varying mask ratio on networks trained on the ImageNet-1K dataset found 75% mask ratios achieved high performance on both finetuning and linear-probing of the encoder's latent space. The MAE processes only unmasked patches during training, increasing the efficiency of data processing in the encoder and lowering the memory usage of the transformer. A less computationally-intensive ViT is used for the decoder in the original implementation of the MAE. Masked patches are added back to the output of the encoder block as mask tokens and both are fed into the decoder. A reconstruction loss is computed for the masked patches to assess network performance. ==== Prediction ==== In prediction, the decoder architecture is discarded entirely. The input image is split into patches by the same algorithm as in training, but no patches are masked out. A linear projection wi

KeyBase

KeyBase is a database and web application for managing and deploying interactive taxonomic keys for plants and animals developed by the Royal Botanic Gardens Victoria. KeyBase provides a medium where pathway keys which were traditionally developed for print and other classical types of media, can be used more effectively in the internet environment. The platform uses a concept called "keys" which can be easily linked together, joined with other keys, or merged into larger other seamless keys groups, with each still available to be browsed independently. Keys in the KeyBase database can be filtered and displayed in a variety of ways, filters, and formats.

Site-specific browser

A site-specific browser (SSB) is a software application dedicated to accessing pages from a single source (site) on a computer network such as the Internet or a private intranet. SSBs typically simplify the more complex functions of a web browser by excluding the menus, toolbars and browser graphical user interface associated with functions that are external to the workings of a single site. Modern site-specific browsers range from simple browser windows without navigation controls to sophisticated desktop applications built with frameworks like Electron that bundle entire browser engines. This evolution has enabled many popular desktop applications to be built using web technologies, effectively making them advanced site-specific browsers. == History == === Early development === One of the earliest examples of an SSB was MacDICT, a Mac OS 9 application that accessed various websites to define, translate, or find synonyms for words typed into a text box. However, the first general-purpose SSB is considered to be Bubbles, which launched in late 2005 on the Windows platform. Bubbles introduced the term "Site Specific Extensions" for SSB userscripts and created the first SSB JavaScript API. In 2007, Mozilla announced Prism (originally called WebRunner), a project to integrate web applications with the desktop. That same year, Todd Ditchendorf, a former Apple Dashboard engineer, released Fluid for macOS. On 2 September 2008, Google Chrome was released with a built-in "Create application shortcut" feature, bringing SSB functionality to mainstream users. This feature allowed any website to be launched in a separate window without the browser interface. === Modern era === The landscape of site-specific browsers changed dramatically with the introduction of Electron in 2013 (originally called Atom Shell). Electron combined Chromium and Node.js into a single runtime, enabling developers to build desktop applications using web technologies. This framework has since powered applications used by hundreds of millions of users, including Visual Studio Code, Slack, Discord, and Microsoft Teams. In 2015, the concept of Progressive Web Apps (PWAs) was introduced by Google engineers Alex Russell and Frances Berriman, representing a parallel evolution in web-to-desktop technology. While PWAs share similar goals with SSBs, they follow web standards and can be installed directly from browsers. More recently, alternative frameworks like Tauri have emerged, offering significantly smaller application sizes by using the system's native web renderer instead of bundling Chromium. == Technical implementation == Site-specific browsers can be implemented through various approaches: === Browser-based SSBs === The simplest form of SSB is created through browser features that allow websites to run in separate windows without the standard browser interface. Modern Chromium-based browsers offer "Install as app" or "Create shortcut" functionality that creates a dedicated window for a specific website. These SSBs share the browser's underlying engine and resources but operate in isolated windows. === Framework-based SSBs === More sophisticated SSBs are built using application frameworks: Electron: Bundles a complete Chromium browser with Node.js, resulting in applications of 85MB or larger. Each Electron application runs its own browser instance, providing full access to system APIs but consuming significant resources. Tauri: Uses the operating system's native web rendering engine (WebView2 on Windows, WebKit on macOS, and WebKitGTK on Linux), resulting in applications typically 2.5-10MB in size. Other frameworks: Include Neutralino.js (ultra-lightweight using system browser), Wails (Go-based), and the Chromium Embedded Framework (CEF). == Comparison with Progressive Web Apps == While site-specific browsers and Progressive Web Apps (PWAs) share the goal of bringing web content to the desktop, they differ in several key aspects: == Applications == Site-specific browsers have become the foundation for many popular desktop applications: Communication and collaboration: Many modern communication tools are built as SSBs, including Slack, Discord, Microsoft Teams, and WhatsApp Desktop. These applications benefit from web-based development while providing desktop integration. Development tools: Visual Studio Code, used by 73.6% of developers according to Stack Overflow's 2024 survey, is built with Electron, as are Atom and GitHub Desktop. Productivity software: Applications like Notion, Obsidian, and various project management tools use SSB technology to provide consistent experiences across platforms. Security and Privacy: Web browsers can be modified to only have access to a single site, in order to protect the security and privacy of the user via compartmentalization == Security and performance == === Memory usage === Framework-based SSBs, particularly those using Electron, are known for high memory consumption. Studies show Electron applications typically use 120-300MB at baseline, with complex applications consuming significantly more. This is approximately 5-10 times more memory than equivalent native applications. === Security considerations === SSBs can provide security benefits through process isolation, where each application runs in its own sandboxed environment. However, bundling an entire browser engine also means each application must be updated independently to patch security vulnerabilities. Research presented at the Network and Distributed System Security (NDSS) Symposium has identified various security challenges specific to Electron applications. === Bundle sizes === The choice of framework significantly impacts application size: Electron applications: 85MB+ (includes full Chromium) Tauri applications: 2.5-10MB (uses system WebView) Browser-based SSBs: No additional download (uses existing browser) == Software == === Browser support === Most modern browsers provide some form of SSB functionality: Chromium-based browsers (Google Chrome, Microsoft Edge, Brave, Opera, Vivaldi): "Install as app" or "Create shortcut" feature Safari: "Add to Dock" feature in macOS Sonoma (2023) Firefox: Removed SSB support in December 2020 (version 85) GNOME Web: "Install Site as Web Application" feature === Standalone tools === ==== Active ==== WebCatalog (Windows, macOS, Linux) – Manages multiple SSBs with isolated storage Fluid (macOS) – Pioneering SSB creator for Mac Unite (macOS) – Creates SSBs with customization options Coherence X (macOS) – Advanced SSB creation tool Pake (cross-platform) – Open-source SSB creator Wavebox (cross-platform) – Workspace browser with SSB features ==== Discontinued ==== Mozilla Prism – Cross-platform SSB creator (discontinued 2011) Nativefier – Command-line SSB creator (discontinued 2023) Epichrome – macOS SSB creator (discontinued 2021) === Development frameworks === Electron – Most popular framework, bundles Chromium and Node.js Tauri – Rust-based framework using system WebView Chromium Embedded Framework (CEF) – C++ library for embedding Chromium Neutralino.js – Lightweight framework using system browser Wails – Go-based framework for web frontends

Color image pipeline

An image pipeline or video pipeline is the set of components commonly used between an image source (such as a camera, a scanner, or the rendering engine in a computer game), and an image renderer (such as a television set, a computer screen, a computer printer or cinema screen), or for performing any intermediate digital image processing consisting of two or more separate processing blocks. An image/video pipeline may be implemented as computer software, in a digital signal processor, on an FPGA, or as fixed-function ASIC. In addition, analog circuits can be used to do many of the same functions. Typical components include image sensor corrections (including debayering or applying a Bayer filter), noise reduction, image scaling, gamma correction, image enhancement, colorspace conversion (between formats such as RGB, YUV or YCbCr), chroma subsampling, framerate conversion, image compression/video compression (such as JPEG), and computer data storage/data transmission. Typical goals of an imaging pipeline may be perceptually pleasing end-results, colorimetric precision, a high degree of flexibility, low cost/low CPU utilization/long battery life, or reduction in bandwidth/file size. Some functions may be algorithmically linear. Mathematically, those elements can be connected in any order without changing the end-result. As digital computers use a finite approximation to numerical computing, this is in practice not true. Other elements may be non-linear or time-variant. For both cases, there is often one or a few sequences of components that makes sense for optimum precision and minimum hardware-cost/CPU-load.

EasyChair

EasyChair is a web-based conference management software system. It has been used since 2002 in the scientific community for tasks such as organising research paper submission and review. In 2012, EasyChair added an open access online publication service for conference proceedings. == Description == EasyChair is a paid web-based conference management software system used, among other tasks, to organize paper submission and review, similar to other event management system software such as OpenConf. EasyChair used to be run by the Department of Computer Science at the University of Manchester but now it is a commercial service, owned by EasyChair Ltd. in Stockport (established 2016). EasyChair used to be free, for standard service, but as of 2022, only minimal services are free. The EasyChair website also provides an open access online publication service for conference proceedings. When launched in 2012, the service was for computer science only, but in 2016 it was expanded to all sciences. == History == The EasyChair software has been in continuous development since 2002. As of 2015, the code base consists of nearly 300,000 lines of code, and it has been used by more than 41,000 conferences. More than two and a half million users in the scientific community reported using it in 2019.

Tertiary review

In software engineering, a tertiary review is a systematic review of systematic reviews. It is also referred to as a tertiary study in the software engineering literature. However, Umbrella review is the term more commonly used in medicine. Kitchenham et al. suggest that methodologically there is no difference between a systematic review and a tertiary review. However, as the software engineering community has started performing tertiary reviews new concerns unique to tertiary reviews have surfaced. These include the challenge of quality assessment of systematic reviews, search validation and the additional risk of double counting. == Examples of Tertiary reviews in software engineering literature == Test quality Machine Learning Test-driven development

Managed private cloud

Managed private cloud (also known as "hosted private cloud" or "single-tenant SaaS") refers to a principle in software architecture where a single instance of the software runs on a server, serves a single client organization (tenant), and is managed by a third party. The third-party provider is responsible for providing the hardware for the server and also for preliminary maintenance. This is in contrast to multitenancy, where multiple client organizations share a single server, or an on-premises deployment, where the client organization hosts its software instance. Managed private clouds also fall under the larger umbrella of cloud computing. == Adoption == The need for private clouds arose due to enterprises requiring a dedicated service and infrastructure for their cloud computing needs, such as for business-critical operations, improved security, and better control over their resources. Managed private cloud adoption is a popular choice among organizations. It has been on the rise due to enterprises requiring a dedicated cloud environment and preferring to avoid having to deal with management, maintenance, or future upgrade costs for the associated infrastructure and services. Such operational costs are unavoidable in on-premises private cloud data centers. == Advantages and challenges of managed private cloud == A managed private cloud cuts down on upkeep costs by outsourcing infrastructure management and maintenance to the managed cloud provider. It is easier to integrate an organization's existing software, services, and applications into a dedicated cloud hosting infrastructure which can be customized to the client's needs instead of a public cloud platform, whose hardware or infrastructure/software platform cannot be individualized to each client. Customers who choose a managed private cloud deployment usually choose them because of their desire for efficient cloud deployment, but also have the need for service customization or integration only available in a single-tenant environment. This chart shows the key benefits of the different types of deployments, and shows the overlap between these cloud solutions. This chart shows key drawbacks. Since deployments are done in a single-tenant environment, it is usually cost-prohibitive for small and medium-sized businesses. While server upkeep and maintenance are handled by the service provider, including network management and security, the client is charged for all such services. It is up to the potential client to determine if a managed private cloud solution aligns with their business objectives and budget. While the service provider maintains the upkeep of servers, network, and platform infrastructure, sensitive data is typically not stored on managed private clouds as it may leave business-critical information prone to breaches via third-party attacks on the cloud service provider. Common customizations and integrations include: Active Directory Single Sign-on Learning Management Systems Video Teleconferencing == Deployment strategies and service providers == Software companies have taken a variety of strategies in the Managed Private Cloud realm. Some software organizations have provided managed private cloud options internally, such as Microsoft. Companies that offer an on-premises deployment option, by definition, enable third-party companies to market Managed Private Cloud solutions. A few managed private cloud service providers are: Adobe Connect: Adobe Connect may be purchased for on-premises deployment, multi-tenant hosted deployment, managed private cloud as ACMS, or managed by third-party managed private cloud provider ConnectSolutions. Rackspace CenturyLink Microsoft licenses for Lync, SharePoint and Exchange may be purchased for on-premises deployment, a multi-tenant hosted deployment via Office 365, or managed by third-party cloud hosting from Azaleos, ConnectSolutions and others.