# Introduction Source: https://docs.pixldata.com/api-reference/introduction Soon. # Audio Annotation Module Source: https://docs.pixldata.com/guides/annotation-modules/audio-annotation-tool # Image Annotation Module Source: https://docs.pixldata.com/guides/annotation-modules/image-annotation-tool # RLHF Module Source: https://docs.pixldata.com/guides/annotation-modules/rlhf-tool # Video Annotation Module Source: https://docs.pixldata.com/guides/annotation-modules/video-annotation-tool # AI Assisted Annotations Source: https://docs.pixldata.com/guides/basics/ai-assisted-annotation PixlHub integrates advanced machine learning models directly into the annotation workflow to accelerate the labeling process. These tools allow for complex annotations to be generated with minimal human input while maintaining high precision. ### **Local Inference & Privacy** All AI models within PixlHub are modified for optimized performance and run **locally**. The platform does not send data to external APIs or third-party servers. This ensures that even when using AI-assisted tools, sensitive data remains entirely within the local environment. ### **Smart AI Tools** The following models are accessible via the **Smart Tool Palette**: **1. Polygon from BBox** This tool combines the speed of a bounding box with the precision of a polygon. * **Workflow:** The user draws a standard box around an object. * **AI Action:** The model instantly identifies the object's boundaries within that box and converts it into a high-fidelity **Polygon**. This is ideal for quickly segmenting objects with complex edges. Box To Polygon **2. Auto-Detect Objects (Local YOLO)** Designed for high-volume initial labeling, this tool can populate an entire image with annotations in a single click. * **Workflow:** Activate the tool with one click on the canvas. * **AI Action:** Utilizing a modified **YOLO** model, the platform detects all recognizable objects in the frame and automatically labels them as **Bounding Boxes** based on the selected label schema. Auto Object Detection Pixlhub **3. Smart Segment (Local SAM)** This tool allows for complex segmentation with zero manual tracing. * **Workflow:** The user places a single **Point** on the target object. * **AI Action:** Powered by a modified **Segment Anything Model (SAM)**, the AI identifies the entire structure of the object and wraps it in a precise **Polygon**. It is the most efficient way to label irregular shapes with a single click. Octopus Polygon From Point ### **Workflow Integration** AI-generated labels are treated as standard annotations. Once the AI provides a prediction, annotators can manually refine, resize, or delete the results just like any other label. This "Human-in-the-loop" approach ensures that speed never comes at the expense of accuracy. *** # Bounding Box Source: https://docs.pixldata.com/guides/basics/bounding-box The Bounding Box is the fundamental tool for object detection and localization. In PixlHub, the Bounding Box is not a static tool; it can be adapted into three distinct variations depending on the project's requirements. Bounding Box Tool ### **The Interactive Toolbar** Upon selecting the Bounding Box from the **Smart Tool Palette**, the **Interactive Toolbar** (located just below the header) dynamically updates. While this toolbar typically houses general controls like zoom and fit-to-screen, it transforms to show specific modification modes whenever a drawing tool is active. This allows users to switch between different BBox behaviors on the fly without changing the underlying label schema. ### **Bounding Box Variations** PixlHub supports three primary modes for box-based annotations: 1. **Standard BBox:** The classic axis-aligned rectangular box used for general object detection. It follows standard industry behavior for resizing and placement. 2. **Rotating BBox:** Accessible via the Interactive Toolbar, this mode allows the box to be rotated at any angle. This is essential for labeling objects that are not aligned with the horizontal or vertical axes (e.g., tilted vehicles or overhead aerial imagery). 3. **OCR Box:** When this mode is active, the selected area is used to extract text content. The OCR Box automatically attempts to identify and digitize any text within the boundaries, linking the string data directly to the annotation. ### **Interaction & Context Menu** Beyond the toolbar, Bounding Boxes include standard interactive features to maintain a high-speed workflow: * **Context Menu:** Right-clicking any BBox on the canvas opens a context menu. This provides quick access to common actions such as deleting, locking, or duplicating the label. * **Canvas Controls:** Standard drag-and-drop placement and corner-point resizing are available across all three BBox variations. # Canvas Settings Source: https://docs.pixldata.com/guides/basics/canvas-settings # Class Definitions Source: https://docs.pixldata.com/guides/basics/class-definitions # Comment Source: https://docs.pixldata.com/guides/basics/comment The Comment Tool is the primary communication layer within PixlHub. It is used to flag edge cases, discuss problematic assets, or provide specific feedback during the labeling and review process. Comment Tool 1 ### **Adding Comments** There are two ways to initiate a conversation within a task: 1. **Coordinate-Based Comments:** Select the Comment Tool from the **Smart Tool Palette** and click on a specific point on the canvas. This opens a modal window where the user can type their message and save it by clicking **"Add Comment."** 2. **Annotation-Specific Comments:** A comment can be linked directly to an existing label (e.g., a specific Bounding Box or Polygon). By right-clicking an annotation and selecting "Add Comment" from the **Context Menu**, the platform automatically embeds the **Annotation ID** into the message, ensuring there is no confusion about which object is being discussed. Comment Tool 2 ### **Collaboration Features** * **Mentions:** Users can use the **"@"** symbol to mention and notify specific team members within a comment. This is the standard way to assign a query to a manager or a fellow annotator. * **Centralized Tracking:** All comments associated with a project are aggregated under the **Comments Tab** in the **Project Details** page. This provides a high-level view of all ongoing discussions and unresolved issues across the entire dataset. Comment Tool 3 ### **Managing Comments** To maintain a clean workflow, comments follow a set of permission rules: * **Resolving & Deletion:** A comment can be marked as "Resolved" once the issue is addressed. * **Permissions:** Standard users can only delete or resolve the comments they have created. **Project Managers**, however, have full administrative rights to resolve or delete any comment within the project to ensure the workflow remains unblocked. # Freeform Source: https://docs.pixldata.com/guides/basics/freeform The Freeform Tool provides a high-speed alternative to the standard Polygon Tool, allowing users to trace complex shapes with a continuous fluid motion rather than manual vertex placement. Freeform Tool ### **Usage and Conversion** * **Drawing Process:** Unlike the click-by-click method of the Polygon Tool, the Freeform Tool allows users to draw shapes freely as if using a pen. * **Auto-Conversion:** Once the stroke is completed, the platform utilizes a path simplification algorithm to automatically convert the freehand drawing into a structured **Polygon**. This ensures that the final annotation remains editable and compatible with standard segmentation formats. * **Finalizing:** Releasing the mouse button or closing the loop automatically triggers the conversion process. ### **Interactive Toolbar: Drawing Modes** The Freeform Tool shares the same operational flexibility as the Polygon Tool. Through the **Interactive Toolbar**, users can select from four distinct modes: 1. **Overwrite:** The new shape is drawn independently. 2. **Subtract:** The freehand stroke cuts out a section from an existing polygon. 3. **Auto Subtract:** The platform automatically trims the freeform shape so it does not overlap with existing boundaries. 4. **Union:** The new freeform shape is merged into a single logical entity with an existing polygon. ### **Customization: Point Density** The precision of the freeform-to-polygon conversion can be adjusted to match the complexity of the object: * **Point Frequency:** Users can control how many vertices are generated during the conversion. A higher density captures more detail, while a lower density creates a simpler, more manageable polygon. * **Configuration:** These settings are located under the **Freeform Settings** section within the **Canvas Settings** menu. Adjusting the frequency allows for a perfect balance between speed and geometric accuracy. # Global Controls Source: https://docs.pixldata.com/guides/basics/global-controls The top navigation bar provides quick access to workspace management, notifications, and personal preferences. These controls are persistent across all pages, ensuring seamless transitions between tasks. Ekran Kaydı2026 01 1215 10 35 ### Organization Switcher Users belonging to multiple organizations can switch between different workspaces using the dropdown menu in the top navigation. Selecting an organization instantly updates the project lists, team members, and datasets associated with that specific entity. ### **Notification System** PixlHub uses a tiered notification system to keep users updated on project status and organizational changes: * **Organization Alerts:** A small **orange badge** appears on the user avatar when there is a critical update or invite at the organization level. * **General Notifications:** Activity updates, such as task assignments or review feedback, are accessible via the notification bell icon. ### **Language & Theme Toggles** To ensure comfort during long working hours, the interface can be customized instantly: * **Dark Mode Support:** Users can toggle between Light and Dark themes. Dark mode is optimized to reduce eye strain during intensive annotation sessions. * **Language Switcher:** The platform supports multiple languages. Switching the language updates all interface elements, menus, and system messages. ### **User Settings & Profile** The user profile menu provides access to account-level configurations. Here, users can manage their personal information, update security settings, and view their individual performance statistics across different projects. # Guidelines Source: https://docs.pixldata.com/guides/basics/guidelines Guidelines ensure that every annotator follows the same rules, maintaining consistency and quality across the entire project. PixlHub allows managers to provide these instructions directly within the project settings and the labeling interface. ### Assigning Guidelines to a Project To add or update instructions for a specific project: 1. **Select the Project:** Navigate to the **Projects** page and click on the relevant project. 2. **Open Settings:** Inside the project dashboard, go to the **Settings** tab. Guideline 1 3. **Locate Annotation Guidelines:** Find the "Annotation Guidelines" row. Clicking this row will expand the section to reveal configuration options. Guideline 2 4. **Choose a Format:** * **PDF Guidelines:** A PDF file can be uploaded directly to provide detailed visual instructions and edge-case examples. * **Text Guidelines:** Alternatively, instructions can be entered manually in the text field for quick, searchable reference. ### **Viewing Guidelines During Labeling** Once assigned, guidelines are immediately available to annotators without them having to leave the labeling environment: Guideline 4 * **Accessing the Guide:** While inside the **Annotation Tool**, users can click the **Guidelines** tab located in the **Right Drawer**. * **On-the-fly Reference:** This allows annotators to double-check labeling rules, class definitions, or specific object boundaries while they are actively working on a task. # Interactive Toolbar Source: https://docs.pixldata.com/guides/basics/interactive-toolbar # Line Source: https://docs.pixldata.com/guides/basics/line The Line Tool is designed for annotating linear structures and boundaries that do not require closed shapes. It is an essential tool for projects involving lane detection, wire tracking, or skeletal mapping. ### **Usage** * **Tool Selection:** Like other annotation tools, the Line Tool is accessed via the **Smart Palette**. * **Drawing:** Users can place consecutive points on the canvas to define the path of the line. The interface maintains high precision for both straight and multi-segmented lines. * **Finalizing:** To complete a line segment, users can press the **Enter** key or double-click the final point. ### **Common Use Cases** While simple in nature, the Line Tool is frequently used for: * **Road Markings:** Defining lanes and dividers in autonomous driving datasets. * **Boundaries:** Marking fences, walls, or architectural edges. * **Skeletonization:** Connecting keypoints to represent human or animal structures. ### **Flexibility** As part of the **Smart Annotation Palette**, the Line Tool can be used interchangeably with other tools like Bounding Boxes or Polygons within the same project. This allows for comprehensive labeling of complex scenes where different geometry types are required. # Object Instances Source: https://docs.pixldata.com/guides/basics/object-instances # Point Source: https://docs.pixldata.com/guides/basics/point The Point Tool is used to mark specific coordinates or landmarks on the canvas. It is an essential tool for tasks requiring high spatial precision, such as keypoint detection, facial landmarking, or pose estimation. Point Tool ### **Usage** * **Tool Selection:** The Point Tool can be activated via the **Smart Palette**. * **Placement:** Users can place a point by simply clicking on the target coordinate on the canvas. Each point acts as an independent annotation linked to the defined label schema. * **Precision:** The tool is designed to provide pixel-perfect accuracy, ensuring that the landmark is placed exactly where the user intends. ### **Customization: Point Size** To accommodate different image resolutions and visual requirements, the visibility of points can be adjusted: * **Adjustable Size:** The diameter of the points on the canvas is not fixed. Users can increase or decrease the point size to ensure they are clearly visible or to avoid obscuring fine details of the underlying asset. * **How to Change:** Point size settings are located under the **Canvas Settings** menu within the annotation interface. Adjusting this setting updates all points on the canvas in real-time. ### **Common Use Cases** The Point Tool is the primary choice for: * **Facial Landmarks:** Marking specific features like eyes, nose, or mouth corners. * **Human Pose Estimation:** Identifying joints and skeletal keypoints. * **Small Object Localization:** Pinpointing small or distant objects where a bounding box might be unnecessary. # Polygon Source: https://docs.pixldata.com/guides/basics/polygon The Polygon Tool is used for high-precision instance and semantic segmentation. It allows for the annotation of complex or irregular shapes that cannot be accurately captured by standard bounding boxes. Polygon Tool 1 ### **Getting Started** * **Tool Selection:** The Polygon Tool is selected via the **Smart Tool Palette**. * **Efficiency (Tool Persistence):** To speed up the workflow, the last selected tool remains active by default. There is no need to re-select the Polygon Tool after completing a shape; the user can immediately begin drawing the next object. * **Closing a Shape:** A polygon can be finalized in two ways: 1. Clicking back on the **starting point** (the first node). 2. Pressing the **Enter** key to automatically close the shape from the last point to the first. ### **Interactive Toolbar: Polygon Modes** When the Polygon Tool is active, the **Interactive Toolbar** provides four specialized modes to handle overlapping shapes and complex geometries. These modes allow users to manage how new polygons interact with existing ones: 1. **Overwrite:** The default mode. New polygons are drawn independently, regardless of existing annotations. 2. **Subtract:** Used to cut out a section from an existing polygon. This is ideal for creating "holes" or removing specific parts from a larger shape (e.g., labeling a donut shape). 3. **Auto Subtract:** A smart mode that automatically trims the new polygon so it does not overlap with existing shapes, ensuring perfectly aligned boundaries without manual effort. 4. **Union:** Allows for the merging of two or more polygons. When used, the new shape is combined with a selected existing polygon to form a single, unified logical entity. ### **Precision & Speed** The tool is optimized for high-speed vertex placement. Whether defining a simple triangle or a complex 100-point boundary, the canvas remains responsive, allowing for professional-grade segmentation at scale. # Right Drawer Source: https://docs.pixldata.com/guides/basics/right-drawer # OCR Tool Source: https://docs.pixldata.com/guides/basics/smart-annotation-palette/ocr-tool The OCR (Optical Character Recognition) Tool is a specialized extension of the Bounding Box, designed to detect and digitize text directly from images. By integrating character recognition into the standard annotation flow, PixlHub eliminates the need for manual transcription. Ocr Boundingbox Tool ### **Activation & Setup** The OCR functionality is linked to the Bounding Box tool. To enable it: 1. **Select BBox:** Choose the Bounding Box from the **Smart Tool Palette**. 2. **Toggle OCR Mode:** In the **Interactive Toolbar** (at the top), switch the mode from a standard BBox to an **OCR Box**. Once active, any box drawn on the canvas is treated as a text detection region. ### **Scanning Modes** PixlHub provides two ways to process text, allowing for both manual control and high-volume automation: * **Manual Scan:** After drawing a box, right-click to open the **Context Menu** and select **"Scan"**. This triggers the recognition engine for that specific region. * **Auto-Scan:** For faster workflows, the **Auto-Scan** option can be toggled on via the Interactive Toolbar. When enabled, PixlHub automatically performs text extraction the moment a box is completed, instantly populating the associated text field. ### **Privacy & Security (Local Inference)** To meet the requirements of high-security environments and sensitive data projects, the OCR engine is built with a "security-first" architecture: * **Local Models:** All recognition is performed using modified OCR models running locally on the system. * **No External Dependencies:** The process does not rely on third-party APIs or external cloud connections. * **Data Sovereignty:** Since no data is transmitted to external servers, the OCR tool is fully compliant with strict data privacy regulations, ensuring that sensitive information never leaves the local environment. # Smart Tool Palette Source: https://docs.pixldata.com/guides/basics/smart-annotation-palette/smart-tool-palette The Smart Annotation Palette is the central interface for all drawing and interaction tasks on the canvas. Unlike traditional labeling environments that restrict a project to a single tool type, this palette provides a versatile toolkit that supports multiple geometry types within the same workspace. Smart Annotation Tool ### **Dynamic Tool Selection** PixlHub allows the label schema to evolve without constant configuration changes. This means a single project is not limited to just one type of annotation (e.g., Bounding Boxes only); managers and annotators can utilize a variety of tools as the project requirements grow. The palette includes the following essential tools: * **Bounding Box (BBox):** The standard tool for object detection and localization. * **Polygon:** For high-precision instance or semantic segmentation of complex shapes. * **Freeform:** A flexible drawing tool for tracing irregular boundaries manually. * **Line & Point:** Specialized tools for keypoint detection, skeletal structures, or boundary markings. * **Comments:** A collaboration tool to place feedback markers directly onto specific coordinates. * **AI Predictions:** An assistive feature that utilizes pre-trained models to generate label suggestions, significantly reducing manual effort. ### **Operational Flexibility** The primary advantage of the Smart Annotation Palette is its adaptability. New label types can be introduced into an existing project seamlessly. Whether the goal is to add a few keypoints to a bounding box project or to transition into full-scale segmentation, the palette ensures that the workflow remains uninterrupted and the label schema stays flexible. # Workspaces Source: https://docs.pixldata.com/guides/basics/workspaces Workspaces (Organizations) are the highest level of the PixlHub hierarchy. They serve as independent environments where teams, data, and projects are managed collectively. ### **Creating a New Organization** A new organization can be initiated from two primary locations: 1. **Organization Switcher:** Click the **"+ Create Workspace"** option within the organization dropdown in the top navigation bar. 2. **User Profile:** Navigate to the **Organizations** section under the profile menu and click **"Create Organization."** After entering a name and a brief description, the organization is created. To enter the new workspace, select it from the list and click the **"Switch to This"** button. ### **Managing Team Members** Users are added to an organization via email invitations through the **Members** tab on the organization page. **Organizational Roles:** * **Member:** The default role. Members can participate in projects they are specifically invited to but cannot modify organization-level settings. * **Admin:** Admins have full visibility and control over all projects, datasets, and settings within the organization. Because an Admin can modify or delete any resource, this role should be assigned with caution. ### **Project-Level Access** Joining an organization does not automatically grant access to its projects. To maintain data security and focus, users must be invited to specific projects individually. Project-level roles include: * **Labeler:** Authorized to annotate and submit tasks. * **Reviewer:** Authorized to inspect, approve, or reject submitted annotations. * **Manager:** Authorized to manage project settings, ontology, and team assignments. ### **Handling Invitations** When a user is invited to a new organization or project, PixlHub sends a notification to keep them informed: * **Visual Indicators:** A small **orange badge** appears on the user’s avatar as a persistent reminder of pending invitations. * **Accepting Invites:** Users can view and respond to these invitations through the **Notifications** panel. Access to the workspace or project is granted immediately upon acceptance. # Assets Source: https://docs.pixldata.com/guides/essentials/assets Assets are the raw data files—such as images or videos—that form the foundation of any labeling project. The Assets page provides a centralized environment to upload, organize, and monitor the status of these files. Ekran Resmi2026 02 0909 59 34 ### **The Upload Process** Data is imported into the project through a dedicated upload interface. * **Multi-File Selection:** Users can select and upload multiple files simultaneously. * **Upload Constraints:** To ensure stability, each upload session is limited to a maximum of **1000 files** and a total size of **512 MB**. * **Continuous Uploads:** If a dataset exceeds these limits, additional files can be added by initiating a new upload session after the current one completes. New files can be assigned to existing batches to maintain data continuity. ### **Organization with Batches** During the upload process, assets can be assigned to a **Batch**. Batches allow managers to group data logically (e.g., by date, source, or priority), making it easier to distribute work and filter datasets as the project grows. ### **Viewing and Navigation** The platform offers two distinct ways to interact with the dataset: * **Gallery View:** Optimized for visual browsing. Clicking on an asset in this mode opens a high-quality **preview** of the file. * **List View:** Designed for data management. Bulk selection and administrative actions are performed exclusively in this view. * **Filtering:** A batch-based filter is available to quickly isolate specific subsets of the data. ### **Status Indicators** In Gallery View, each asset features a small color-coded indicator in the bottom-right corner to show its current labeling progress: * **Blue Indicator:** The asset is raw and has not yet been labeled. * **Green Indicator:** The asset has been successfully labeled. Pixl Hub Data Labeling Platform 02 09 2026 10 04 AM ### **Data Management and Deletion** To prevent accidental data loss, management actions follow specific rules: * **Individual/Bulk Deletion:** Files can be selected and deleted only while in **List View**. * **Project Deletion:** To remove an entire dataset along with the project, it is recommended to use the **Project Settings** menu rather than deleting files individually. This ensures a clean removal of all associated metadata and tasks. Assets2 # Exports Source: https://docs.pixldata.com/guides/essentials/exports The **Export** page is where users extract their annotated data for machine learning model training, analysis, or third-party integration. PixlHub provides a transparent preview of the dataset and supports a wide array of industry-standard formats to ensure compatibility with any AI pipeline. ### **Export Summary & Analytics** Before generating a file, PixlHub provides a real-time summary of the data included in the current selection. This helps managers verify the dataset's composition without needing to open the files: * **Task Status Counts:** View the exact number of **Completed**, **Labeled**, and **Ready** tasks. * **Annotation Types:** A breakdown of the geometry types included (e.g., Bounding Boxes, Polygons, Points). * **Labels Distribution:** A detailed list showing how many instances of each class (e.g., "Pothole," "Vehicle") are present in the export. ### **Filtering Options** To export specific subsets of data, users can apply a **Date Range Filter**. This allows for the extraction of only those tasks that were completed within a specific timeframe, making it easy to manage versioning or incremental model updates. *** ### **Supported Export Formats** PixlHub supports a diverse range of formats, including a **Schema Live Preview** that shows an example structure of the chosen format before the export starts: * **JSON (Native):** The most comprehensive format. It supports all annotation types and nested attributes. Recommended for backups or migrating data between PixlHub projects. * **COCO JSON:** The industry standard for object detection and segmentation tasks. * **YOLO (Darknet):** Optimized for YOLO-based object detection models. * **YOLOv8 Segmentation:** Specifically formatted for the latest segmentation models in the YOLO ecosystem. * **Pascal VOC (XML):** A classic format widely used in academic and legacy computer vision projects. * **CSV (Simple):** A tabular format for quick data analysis or spreadsheet-based reviews. * **Custom Template:** Allows users to define their own structure to meet specific pipeline requirements. *** ### **Advanced Export Options** Users can fine-tune the contents of the export package using the following configuration toggles: * **Include Asset Metadata:** Adds dimensions, batch information, and other asset-level details to the export. * **Include Non-Reviewed Tasks:** By default, exports often focus on completed data. This option allows including tasks that are **Labeled** but have not yet passed the **Review** stage. * **Include Review Metadata:** Includes information regarding the review process, such as the reviewer's name and the date of approval. * **Organize by Train/Val/Test Split:** Automatically organizes the exported files into separate folders based on the split assignments defined in the **Tasks** page. This makes the data immediately ready for model training scripts. *** ### **Export History** Every export generated is stored in the **Export History** table. This allows team members to download previous versions of the dataset at any time, providing a clear audit trail of what data was used for which model version. # Imports Source: https://docs.pixldata.com/guides/essentials/imports The **Import** page allows you to bring existing annotations from external sources or previous projects into PixlHub. Whether you are migrating a dataset or utilizing pre-labeled data, the import system ensures that labels are correctly mapped to your current assets. ### **Pre-Import Checklist** To ensure a successful data merge, follow these preparation steps: * **Task Status:** Change the status of the target tasks to **"Prepare."** Annotations can only be imported into tasks that are in this stage to prevent overwriting active work. * **Format Selection:** Prepare your annotation files in one of the supported formats: **PixlHub Native (JSON)**, **COCO**, or **YOLO**. ### **Import Workflow** The system follows a four-step verification process to maintain data integrity: 1. **Upload:** Drag and drop or click to upload your annotation file (Native JSON or COCO/YOLO formats). 2. **Review Validation:** PixlHub automatically parses the file and provides a validation report. This shows how many labels were found and how they align with your current project. 3. **Conflict Check:** Review any warnings regarding unmatched filenames or ID discrepancies. 4. **Confirm:** Once the validation results are reviewed, click confirm to apply the annotations to your tasks. ### **Supported Formats & Matching Logic** The accuracy of an import depends on the format and the underlying matching logic: * **PixlHub Native Format (Recommended):** This is the most accurate method. It uses **Unique IDs** to match annotations directly to the specific asset and label schema. It is highly recommended for project migrations or backups. * **External Formats (COCO & YOLO):** These formats rely on **Filename Matching**. While convenient for importing data from other platforms, this method is less precise as it can be affected by duplicate filenames across different folders. > **Note:** For the best accuracy and to avoid issues with duplicate filenames, use the PixlHub Native format whenever possible. ### **Import History** Every import attempt is logged in the **Import History** table, providing a full audit trail of external data additions. This table includes: * **Date & Format:** When the import occurred and which file type was used. * **Status:** Whether the import was successful, failed, or completed with warnings. * **Imported vs. Total:** A comparison of the number of annotations successfully added versus the total found in the file. * **Matched Tasks:** The number of tasks that successfully received new labels. * **Errors:** A detailed count of any issues encountered, allowing for quick troubleshooting of file formatting or naming conflicts. # Label Schema Source: https://docs.pixldata.com/guides/essentials/label-schema The **Label Schema** (or Ontology) is the structural blueprint of a project. It defines the rules, classifications, and metadata requirements for every annotation. PixlHub provides a highly flexible schema builder, allowing for granular data collection through nested hierarchies and diverse input types. ### **Building the Ontology** A project is organized around **Classes** (e.g., "Vehicle," "Sign," "Obstacle"). To capture detailed information without cluttering the class list, each class can be refined with multiple **Attributes**. The schema supports several input types to ensure data is structured correctly: * **Text:** For manual entries like license plates or serial numbers. * **Number:** For numerical values, counts, or measurements. * **Radio:** For selecting a single option from a list. * **Checkbox:** For simple boolean (True/False) properties. * **Dropdown:** For choosing one option from a large list. * **Multi-select:** For applying multiple relevant tags to a single object. Label Schema 1 ### **Nested (Child) Attributes** To manage complex data requirements, PixlHub utilizes **Child Attributes**. This feature introduces conditional logic to the annotation process, ensuring that the interface remains clean and focused. * **Conditional Logic:** Child attributes only appear when a specific "Parent" option is selected. For example, selecting "Vehicle Type: Truck" can trigger a sub-menu for "Trailer Type," which would remain hidden if "Sedan" were selected instead. * **UI Optimization:** By hiding irrelevant fields, child attributes reduce cognitive load for annotators, minimizing errors and speeding up the labeling process. * **Deep Hierarchies:** Managers can create multiple levels of nesting to capture high-fidelity data for specialized AI models. ### **Global Attributes** While standard attributes describe specific objects, **Global Attributes** are used to define metadata for the entire asset or task. These are ideal for capturing situational context, such as: * **Environment:** Weather conditions (Rain, Snow), lighting (Day, Night), or location type. * **Data Quality:** Image clarity, sensor noise levels, or "Unlabelable" flags. * **Scene Classification:** Identifying the overall category of the frame (e.g., Urban, Rural, Highway). Labelschema Globalattr ### **Customization & Efficiency** The Label Schema includes features designed to maximize annotator throughput and maintain visual organization: * **Color Assignment:** Each class can be assigned a unique color. This makes the canvas instantly readable and helps distinguish between overlapping objects at a glance. * **Keyboard Shortcuts:** Classes and attributes can be mapped to specific hotkeys (e.g., **1, 2, 3, 4**). This allows annotators to switch between labels and confirm attributes without moving the mouse, significantly reducing the time-per-task. # Tasks Source: https://docs.pixldata.com/guides/essentials/tasks Tasks are the operational units of PixlHub. While **Assets** represent the raw data, **Tasks** represent that data within a workflow, assigned to specific users and tracked through various stages of the labeling lifecycle. ### **Task Management & Metrics** The Tasks page provides a high-level overview of project progress through a detailed data table. Each entry allows managers to monitor the health and velocity of the project using the following metrics: * **Status:** The current stage of the task (e.g., Prepare, Label, Review, Completed). * **Assignee:** The specific user currently responsible for the task. * **Duration:** The active time spent by annotators on the task. * **Object Count:** The total number of annotations/objects created within the task. * **Split:** The dataset category (Train, Validation, or Test) assigned to the task. * **Task History:** A chronological log of every action, status change, and user interaction. * **Operational Tags:** Visibility into priority levels, batch assignments, "skipped" status, and the exact time the task entered the queue. Tasks Columns ### **Bulk Operations** To manage large-scale datasets efficiently, PixlHub allows for administrative actions to be performed on multiple selected tasks simultaneously. #### **1. Change Status** This core operation moves tasks between different stages of the workflow. When changing status, managers can also perform "cleanup" actions via a modal window: * **Remove Annotations:** Wipes all existing labels so annotators can start from scratch. * **Remove Assignees:** Unassigns users, returning the tasks to the general pool/queue. * **Remove Stage History:** Deletes the chronological record of the task's journey. * **Remove Duration:** Resets the "time spent" metric to zero. Ekran Resmi2026 02 0910 23 53 #### **2. Set Split** This function organizes data for machine learning workflows. Tasks can be assigned manually or via **Auto Split** into three standard categories: * **Train:** Data used for model training. * **Validation:** Data used for hyperparameter tuning and progress monitoring. * **Test:** Independent data used for the final model evaluation. Split #### **3. Clear Skip** If an annotator "skips" a task due to ambiguity or poor image quality, it is removed from the active queue. The **Clear Skip** operation resets this status, allowing the task to be re-evaluated or re-labeled. # Teams & Roles Source: https://docs.pixldata.com/guides/essentials/teams-and-roles # Introduction to PixlHub Source: https://docs.pixldata.com/index PixlHub is a high-performance data annotation platform designed to transform raw data into structured, production-ready training sets for AI models. It serves as a unified workspace that bridges the gap between massive datasets and high-precision ground truth. ### The Architecture PixlHub uses a structured hierarchy to maintain order across large-scale data operations: * **Organizations:** The highest administrative level. This is where teams, permissions, and multiple workspaces are managed. * **Projects:** The operational hub for specific labeling goals. Each project contains a unique **Ontology** (classes and attributes) and specialized annotation tools. * **Datasets & Assets:** The library of raw source files. Assets are uploaded once and can be utilized across multiple projects without data duplication. **Tasks:** The actual unit of production. PixlHub separates *Assets* from *Tasks*, allowing for diverse labeling objectives to be generated from the same raw source material.\ \ **Optimized Annotation Flow** The interface is built for speed. With a keyboard-first design and a high-performance canvas, the platform remains fluid even when handling complex polygons or hundreds of objects in a single frame. **Integrated Quality Control** Quality is enforced through a dedicated **Review Workflow**. Managers can track every label in real-time, providing feedback, rejecting errors, or approving completions to ensure the final output meets strict accuracy thresholds. **Operational Insights** Project progress is monitored through live dashboards. PixlHub provides granular analytics on label distribution, annotator velocity, and time-per-task, offering full visibility into the project lifecycle. The fastest way to get up and running. This section covers creating an organization, setting up a first project, and uploading the initial dataset to start labeling in minutes. A deep dive into the PixlHub hierarchy. Learn how Organizations, Projects, Assets, and Tasks interact to keep large-scale data operations organized and scalable. A detailed look at the labeling interface. This guide explores the full suite of tools—from precision bounding boxes and polygons to complex classification attributes and keyboard shortcuts. Understand how to set up review cycles for quality control, track team performance through analytics, and export final annotations in various formats. # Quickstart Guide Source: https://docs.pixldata.com/quickstart This guide provides the essential steps to launch a first labeling project on PixlHub. The process is designed to take less than five minutes from setup to the first annotation. ### 1. Create an Organization The first step is setting up an **Organization**. This acts as the primary workspace where team members are managed and projects are hosted. Once the organization is created, it serves as the central hub for all data operations. Ekran Resmi2026 01 0913 01 44 ### 2. Setup a Project Inside the organization, create a new project to define the workspace: * **Tool Type:** Choose the primary format (Image, Video, or Text). * **Ontology:** Define the classes (e.g., "Vehicle") and attributes (e.g., "Color") that will be used for labeling. Ekran Resmi2026 01 0913 03 05 ### 3. Import Data (Assets to Tasks) PixlHub simplifies the workflow by automatically converting uploaded files into actionable tasks. * **Default Import:** When raw files (**Assets**) are uploaded, PixlHub automatically creates tasks for them and places them in the **Label** status. * **Prelabel Import:** If existing annotations are being migrated into PixlHub, they can be imported into the **Prelabel** status for further refinement. ### 4. Start Labeling There is no need to manually assign work. When a user clicks **"Start Labeling"** on the project dashboard, PixlHub automatically assigns the next available task from the queue to that user. ### 5. Task Lifecycle Each task moves through a specific lifecycle based on its current progress. Understanding these four statuses is key to managing the project: * **Prelabel:** Reserved for imported data that already contains initial annotations. * **Label:** The default starting point. Tasks in this status are in the queue, waiting to be picked up and labeled by a user. * **Review:** Once a user submits their work, the task moves to this status. It indicates the task is labeled but requires manager approval. * **Complete:** Tasks that have passed the review process and are finalized. These are ready for export. **Need help?** See our documentation or join our Slack channel.