Changelog

Confusion matrix CSV for QA, more reliable uploads, faster YOLO Classification imports, stable 3D annotation downloads, and key security fixes including an OPA upgrade.NewDownload quality report as a confusion matrix (CSV)‍You can now export quality reports as a confusion matrix CSV. This makes it easy to analyze misclassifications in spreadsheets, share results with stakeholders, and feed downstream QA dashboards. FixesMore reliable uploads and task creationTUS uploads now read HTTP request data in chunks instead of loading the full body into memory. This reduces memory pressure during large uploads. After a TUS upload completes, HEAD requests now return 404 instead of 500. This prevents false alarms in CI and health checks.When creating a task, manifest files uploaded with videos are no longer ignored, so media and metadata stay in sync.Faster imports and stable exportsUltralytics YOLO Classification imports are faster, improving iteration speed for classification workflows.Fixed an AssertionError during 3D annotation download, restoring reliable export of 3D work.Smoother annotation UIRemoved redundant PATCH /api/jobs/<id>/meta calls from the annotation view to cut unnecessary traffic and reduce noise.CLI agent resilienceFixed agents waiting an inappropriate amount of time between retries when acquiring AR tasks. (#10193)Agents now handle failures of queue-related HTTP requests more robustly, improving long-running automation. (#10194)SecurityFixed XSS vulnerabilities related to skeleton SVG images. Upgraded OPA to 1.12.2, aligning with current policy-engine improvements. Staff users can no longer change their own permissions, closing a privilege escalation gap.Have suggestions or requests for what you'd like to see next? Open an issue on GitHub or contact us via Help Desk.

We’re excited to announce another special addition to our automatic and model-assisted labeling suite: Hugging Face Transformers.Hugging Face Transformers is an open-source Python library that provides ready-to-use implementations of modern machine learning models for natural language processing (NLP), computer vision, audio, and multimodal tasks.The library includes thousands of pre-trained models, including a broad selection of computer vision models that you can now connect to CVAT Online and CVAT Enterprise for automated data annotation.The current integration supports the following tasks:Image classificationObject detectionObject segmentationAll you need to do is pick a model you want to label your dataset with from Transformers library, connect it to CVAT via the agent, run the agent, and get fully labeled frames or even entire datasets, complete with the right shapes and attributes in a fraction of the time.Annotation possibilities unlockedJust like with Ultralytics YOLO and Segment Anything Model 2 integrations, this addition opens up multiple workflow optimization and automation opportunities for ML and AI teams.(1) Pre-label data using the right model for the taskConnect any supported Hugging Face Transformers model that matches your annotation goals—whether it’s a classifier, detector, or segmentation model—and run it directly in CVAT to pre-label your data. Each model can be triggered individually, enabling you to generate different types of annotations for the same dataset without scripts or external tools.(2) Label entire tasks in bulkWorking with a large dataset? Apply a model to an entire task in one step. Open the Actions menu and select Automatic annotation. CVAT will send the request to your agent and automatically annotate all frames across all jobs, reducing manual effort and repetitive work.(3) Share models across teams and projectsRegister a model once and make it instantly available across your organization in CVAT. Team members can use it in their own tasks with no local setup, ensuring consistent labeling workflows at scale.(4) Validate model performance on real dataEvaluate any custom or fine-tuned Hugging Face Transformers model directly on annotated datasets in CVAT. Compare model predictions with human labels side-by-side, identify mismatches, and spot edge cases—all within the same environment.How it worksStep 1. Register the functionCreate a native Python function that loads your Hugging Face model (e.g., ViT, DETR, or segmentation transformers) and defines how predictions are returned to CVAT. Register this function via the CLI.Note: The same function works for both CLI-based and agent-based annotation.Step 2. Start the agentLaunch an agent through the CLI. It connects to your CVAT instance, listens for annotation requests, runs your model, and returns predictions back to CVAT.Step 3. Create or select a task in CVATUpload your images or video and define the labels, depending on your evaluation needs and model output.Step 4. Choose the model in the UIOpen the AI Tools panel inside your job and select your registered Hugging Face model under the corresponding tab.Step 5. Run AI annotationCVAT sends the request to the agent, which performs inference and delivers predictions back in the form of annotation shapes tied to the correct label IDs.Get startedReady to enhance your annotation workflow with Hugging Face Transformers? Sign in to CVAT Online and try it out.For more information about Hugging Face Transformers, visit the official documentation.For more details on CVAT AI annotation agents, read our setup guide.

Note: ‍This is the first part of a three-step integration. At the moment, SAM 3 is available only for segmentation tasks in a visual-prompt mode (clicks/boxes), not via text prompts. Free tiers (Community and Online Free) get demo-mode access to SAM 3 suitable for evaluation, not high-volume labeling.For regular production labeling at scale, we recommend using SAM 3 from Online Solo, Team, or Enterprise editions where it’s part of the standard AI tools offering.‍We’re excited to announce the integration of Meta’s new Segment Anything Model 3 (SAM 3) for images and videos segmentation in CVAT.Since we introduced SAM in 2023, it has quickly become one of the most popular methods for interactive segmentation and tracking, helping teams label complex data much faster and with fewer clicks.SAM 3 takes this even further by introducing a completely new segmentation approach, and we couldn’t wait to bring it to CVAT as soon as it was publicly released. The current SAM 3 integration is already available across all editions of CVAT and continues our commitment to bringing state-of-the-art AI tools into your annotation workflows.What’s New in SAM 3?Released in November 2025, SAM 3 is not just a better version of SAM 2, it’s a new foundation model built for promptable concept segmentation. Unlike its predecessor, which segments (and tracks) specific objects you explicitly indicate with clicks, SAM 3 can detect, segment, and track all instances of a visual concept in images and videos using text prompts or exemplar/image prompts.Say you want to label all fish in an underwater video dataset. With SAM 2, you would typically have to initialize the object manually (for example, click or outline fish one by one), and then keep re-initializing or correcting the model whenever new fish appear, fish overlap, or the scene changes. In other words, the workflow is object-by-object and sometimes frame-by-frame. While tracking can propagate the masks or polygons to subsequent frames, every new object or missed instance still requires manual initialization.With SAM 3, the goal is closer to concept-first labeling: you provide a prompt that describes the concept (like “fish”) or show an example, and the model attempts to find and segment all matching instances across the video, and then track them as the video progresses.This shift toward concept-first segmentation opens up new possibilities for automated data annotation workflows. Instead of repeatedly initializing individual objects with clicks or boxes, annotators can focus on defining what needs to be labeled, while the model handles identifying matching instances. This can significantly reduce manual effort on large or visually complex datasets.Key Capabilities of SAM 3Concept-level promptsSAM 3 takes short text phrases, image exemplars, or both, and returns masks with identities for all matching objects, not just one instance per prompt.Unified image + video segmentation and trackingA single model handles detection, segmentation and tracking, reusing a shared perception encoder for both images and videos.Better open-vocabulary performanceOn the new SA-Co benchmark (“Segment Anything with Concepts”), SAM 3 reaches roughly 2× better performance than prior systems on promptable concept segmentation, while also improving SAM 2’s interactive segmentation quality.Massive concept coverageSAM 3 is trained on the SA-Co dataset with millions of images and videos and over 4M unique noun phrases, giving it wide coverage of long-tail concepts.Open-source releaseMeta provides code, weights and example notebooks for inference and fine tuning in the official SAM 3 GitHub repo.SAM 3 for Image Segmentation in CVATNote: This is the first part of a three-step integration. At the moment, SAM 3 is available only for segmentation tasks in a visual-prompt mode (clicks/boxes), not via text prompts. Free tiers (Community and Online Free) get demo-mode access to SAM 3 suitable for evaluation, not high-volume labeling.In the current integration, CVAT exposes the visual side of SAM 3 with point/box based interactive segmentation, because that’s what fits naturally into the existing AI tools UX and doesn’t force you to change your labeling pipelines overnight.Text prompts, open-vocabulary queries, and SAM 3’s native video tracking API are not wired into the UI yet, so it doesn’t behave as a full concept-search engine. Yet, now that it's in CVAT, it remains a very strong interactive segmentation tool compared to other deep learning models, including its predecessor. So, while our engineering team is working on adding the textual prompts annotation, our in-house labeling team decided to test drive SAM 3 labeling capabilities versus SAM 2 on real annotation tasks, and see in which use cases and scenarios each model performs best.SAM 3 vs. SAM 2 Head-to-Head TestTo understand how SAM 3 performs in real labeling workflows, our in-house labeling team compared it with SAM 2 on 18 images covering different object types, sizes, textures, colors, and scene complexity. Both models were tested using interactive visual prompts (points and boxes), with and without refinement.As expected, the results show that there is no single “better” model — each performs best in different scenarios.Where SAM 2 still performs betterSAM 2 tends to produce cleaner, more stable masks with fewer edge artifacts when:Working with simple to medium-complexity objectsObjects have clear, well-defined boundaries and stable shapesSmooth, clean edges are importantAnnotating people on complex backgroundsMinimal refinement and predictable behavior are requiredWhere SAM 3 shows advantagesSAM 3 starts to outperform SAM 2 in more challenging conditions, such as:Complex scenes with many objects, noise, or motion blurObjects where a fast initial shape is more important than perfect boundariesSmall, dense, or touching objects (for example, bacteria)Low-contrast imagery or objects with subtle visual cues, such as soft or ambiguous boundariesWhere both models perform similarlyIn many common cases, both models deliver comparable results:Simple, high-contrast objectsLarge numbers of similar objects annotated individually (for example, grains)Common objects such as flowers, berries, or sports balls when pixel-perfect accuracy is not requiredKey takeawayThere is no universal winner here. At least in the current integration setup. SAM 2 is more stable and predictable, especially around boundaries, while SAM 3 is more flexible and often better suited for complex scenes and hard-to-separate objects. In practice, the best results come from having both tools available and choosing based on the specific task.Get Started with SAM 3 in CVATTo try SAM 3 in CVAT:Create a segmentation task (images or video frames).Open a job in the CVAT Editor.In the right panel, go to AI tools → Interactors.Select Segment Anything Model 3.Use positive and negative clicks or boxes to generate a mask and accept the result.If needed, convert the mask to a polygon and refine it manually.SAM 3 is available in all CVAT editions. In Community and Online Free plans, it runs in demo mode for evaluation purposes.What’s NextThis post covers the first stage of the SAM 3 integration in CVAT: interactive image segmentation. Looking ahead, SAM 3 opens up several directions we’re actively working toward:Text-driven object discovery and pre-labelingMore advanced video object tracking built on SAM 3’s internal tracking capabilitiesWe’ll introduce these features incrementally and share updates as they become stable and ready for production annotation workflows.For now, we anchorage you to try SAM 3 in your next segmentation task and compare it with SAM 2 on your own data.Have questions or feedback? Please reach out via our Help Desk or open an issue on GitHub. Your input helps shape the next steps of the integration.

Instead of our usual month-by-month roundup, this final Digest of 2025 looks back at the core product features and notable CVAT moments we shipped and celebrated this year.From agentic automation to faster video tracking, stronger QA, better analytics, safer API access, and smoother 3D workflows—here’s the 2025 highlight reel.#1 Agentic automation: AI Agents and ML models integration2025 marked a shift in how automation works in CVAT. Instead of treating ML as an add-on, CVAT introduced AI Agents—a way to plug your own models directly into annotation workflows.AI Agents allow teams to run auto-annotation using models hosted on their own infrastructure, while keeping annotators inside CVAT. This makes it easier to automate labeling, reuse existing ML pipelines, and scale annotation without changing tools.We also shared practical examples of how AI Agents work with popular model ecosystems, including SAM 2 and Ultralytics YOLO, showing how agentic labeling fits into real-world workflows.Related posts:Announcing CVAT AI AgentsCVAT AI Agents: What's NewUltralytics YOLO Agentic Labeling#2 SAM 2 for automated video object trackingVideo annotation saw a major upgrade in 2025 with the introduction of SAM 2–based tracking.For CVAT Enterprise, SAM 2 is available as a native tracking solution designed for production-scale video annotation. It significantly reduces manual work when tracking objects across frames and is optimized for large video datasets.For CVAT Online users, SAM 2 tracking is available via AI Agents. This gives teams flexibility to run the tracking model in their infrastructure while keeping annotation workflows centralized in CVAT.Related posts:SAM 2 for object tracking in videos (Enterprise)SAM 2 tracking via AI Agents (Online)#3 Faster QA with honeypots and immediate job feedbackAs annotation teams scale, quality control needs to happen continuously, not only at the final review stage. In 2025, CVAT introduced features that help teams detect issues earlier and reduce rework.Honeypots use pre-annotated reference data mixed into regular jobs to automatically evaluate annotation quality. This makes it possible to monitor quality across the pipeline using a limited but reliable set of ground-truth samples.Immediate Job Feedback complements this by providing quality insights right after a job is completed. Annotators can fix issues immediately, without waiting for delayed review cycles.Related posts:Annotation QA with honeypots‍Immediate Job Feedback#4 Advanced annotation analytics for teams who want visibilityTo give teams better visibility into how work gets done, CVAT introduced advanced analytics. It provides metrics such as working time, annotation speed, object counts, label usage, and productivity trends. Reports can be refreshed when needed, making it easier to monitor performance, identify bottlenecks, and plan resources.Related post:Advanced annotation analytics#5 Safer automation with Personal Access Tokens for API/SDK/CLIAutomation is a core CVAT use case, and in 2025 we made it more secure. Personal Access Tokens (PATs) are a modern way to authenticate to CVAT’s API/SDK/CLI without reusing passwords or relying on legacy keys. You can scope permissions, set expirations, and revoke tokens without disrupting other integrations.Related post:Personal Access Tokens#6 UX and usability improvementsAlongside major features, 2025 included a series of usability improvements aimed at saving time in everyday work.Bulk actions make it possible to manage multiple resources at once, reducing repetitive operations in large workspaces.Project copying workflows are now supported through project backup and restore, allowing teams to reuse configurations, templates, and structures across projects.Keyboard shortcut customization gives power users full control over shortcuts, with support for global and workspace-specific settings and built-in conflict detection.Related posts:Bulk actions documentation‍Keyboard shortcut customization#7 3D / point cloud annotation improvementsPoint cloud annotation workflows received focused improvements in 2025, aimed at stability and precision during long annotation sessions.Updates included smoother navigation, more predictable zoom behavior, improved object focusing, better control point handling, and more stable camera behavior across views.Related post:Point cloud annotation updates#8 CVAT Academy launchIn 2025, CVAT launched CVAT Academy, a dedicated learning space for anyone who wants to build annotation skills faster.CVAT Academy offers hands-on training covering core annotation concepts as well as more advanced workflows, helping individuals and teams onboard more efficiently and work more consistently.Explore CVAT Academy‍And that's a wrap on 2025. Thank you for choosing CVAT for your dataset labeling this year. Whether you use CVAT Online or CVAT Enterprise in production, contribute to the CVAT Community open-source edition, or build your own workflows on top of CVAT, thank you for being with us.See you in 2026!

Point cloud annotation is fundamentally more demanding and complex than 2D labeling. Annotators have to work in sparse, multi-view environments, constantly switch perspectives, and maintain spatial context while dealing with small or partially occluded objects. Our labeling services team knows better than anyone that even minor issues in navigation, zoom behavior, or camera stability can significantly slow down the workflow.That’s why, in our recent release, we’ve focused on practical updates that remove everyday friction and make 3D annotation faster, smoother, and more predictable.#1 Quick object focus with double-clickFinding and focusing on objects in 3D space used to require manual camera rotation and careful positioning. Now, a simple double-click on any object automatically centers the camera on it. This seemingly small change dramatically speeds up annotation, especially when working with smaller objects that are difficult to locate manually. The feature works consistently across both 3D and 2D annotation tasks.#2 Consistent zoom behavior across modes Previously, zoom settings were inconsistent across different annotation modes, requiring constant readjustment. We've standardized the Attribute annotation zoom margin setting to work uniformly across all annotation modes, including 3D side view projections. Even better, zoom levels now persist when switching between objects—no more losing your carefully adjusted view every time you move to the next annotation.#3 Stable and configurable control pointsControl points are essential for precise annotation, but they were behaving erratically in 3D views. They would change size, become transparent, or scale incorrectly with camera distance. We've fixed this completely: the Control point size setting now works properly in 3D, and control points maintain consistent, readable size regardless of camera position or zoom level.#4 Camera stability across projections and view changesSwitching between the main view and side views used to reset the camera position, forcing annotators to re-navigate each time they wanted to check an object from different angles. The camera now maintains its position when moving between projections, allowing smooth multi-angle verification without repetitive navigation.#5 Improved UI readabilityWe've improved the contrast and readability of contextual menus and reduced transparency in areas where it was obscuring text. Small visual improvements like these reduce eye strain and cognitive load during long annotation sessions.#6 Improved zoom limits and touchpad supportWe've increased the maximum zoom level and completely reworked the zoom algorithm to work properly with trackpads. Navigation is now smoother and more predictable, particularly on laptops where trackpad gestures are the primary input method.These updates are available across all CVAT editions: CVAT Online, Enterprise, and Community. Improvements like these come directly from our hands-on work with real customer data and have immediate impact on annotation speed and quality, especially for complex workflows like 3D and point cloud annotation. If you work with point cloud data, we hope these updates will streamline your labeling workflow. Have feedback or suggestions? We'd love to hear from you! Contact our team via HelpDesk or submit an issue on GitHub.

We're excited to announce Personal Access Tokens (PATs), a new and improved authentication method for CVAT's API, SDK, and CLI. If you're building integrations, running automated scripts, or working with CVAT programmatically, this feature is designed to make your workflow more secure and convenient. Why Personal Access Tokens? Until now, authenticating with CVAT's API required using your username and password or legacy API keys. While functional, this approach had some limitations: Security risks: Sharing your password with multiple applications increases exposure if one gets compromised Limited control: No way to restrict what specific applications can do Manual management: Changing your password meant updating it everywhere Personal Access Tokens solve these problems by giving you better control over API access. What Are Personal Access Tokens? Think of PATs as specialized keys for your CVAT account. Instead of using your password in scripts or third-party tools, you create a unique token for each use case. Each token can have: Custom permissions: Choose between read-only or read/write access Expiration dates: Set tokens to automatically expire after a specific time Individual management: Revoke any token instantly without affecting others This means if you suspect a token has been compromised or simply don’t need a specific token anymore, you can revoke just that one token without disrupting your other integrations or changing your password. Key Benefits Enhanced Security: Use separate credentials for each application and eliminate the need to embed passwords in your code. Better Control: Configure exactly what each token can do. Need a script that only reads data? Create a read-only token. Easy Management: Create, edit, and revoke tokens anytime directly from your CVAT user settings. Automatic Cleanup: Unused tokens are automatically removed after a period of inactivity, reducing security risks from forgotten credentials. Getting Started Creating your first Personal Access Token is simple: Navigate to your Profile page Go to the "Security" section Click the "+" button to create a new token Configure the name, expiration date, and permissions Save and securely store the token value (it's only shown once!) Once you have your token, use it in your API requests with the Authorization header: import requests token = "your_token_value" response = requests.get( "https://app.cvat.ai/api/tasks", headers={"Authorization": f"Bearer {token}"} ) Or, if you prefer working with the CLI, set the token as an environment variable: export CVAT_ACCESS_TOKEN="your_token_value" && cvat-cli task ls Important Security Reminders Store tokens securely: Treat them like passwords Set expiration dates: Always configure tokens to expire Use minimal permissions: Grant only the access level needed Revoke immediately: If you suspect a token is compromised, revoke it right away Never share tokens: Each user and application should have its own token Learn More Personal Access Tokens are available now for all CVAT users across Community, Online, and Enterprise editions. We recommend migrating your existing integrations to use PATs for improved security and control. Ready to start using Personal Access Tokens? Check out our complete documentation for detailed instructions on creating, managing, and using PATs in your CVAT workflows. Have questions or feedback? Join the conversation on our Discord or open an issue on our GitHub repository.

Welcome to the October edition of the CVAT Digest, your monthly roundup of the latest features, improvements, and fixes in Computer Vision Annotation Tool (CVAT).This month’s updates make CVAT smoother for teams working with large cloud-based datasets, 3D projects, and automation workflows. Whether you’re managing storage, creating tasks, or connecting external tools via API, you’ll notice a faster, more flexible, and more secure experience across the board.NewToken‑based Authentication Across the PlatformYou can now use API access tokens everywhere:Server API supports API access tokens.CLI accepts tokens via the CVAT_ACCESS_TOKEN environment variable.SDK can authenticate with an API token via login()/make_client().Create Tasks Without LabelsSpin up tasks first and finalize taxonomy later. Great for quick intake from cloud or bulk uploads.Cloud Dataset HandlingCVAT now reliably supports related images for both 2D and 3D tasks from cloud storage, and the Dataset Manifest tool handles 3D datasets across all supported layouts.Admin Control: Max Jobs per TaskSet a maximum number of jobs allowed per task to keep workloads tidy and predictable.Configurable Disk‑Usage Health CheckTune the threshold that triggers the disk‑usage health check to better fit your environment.UpdatesSecurity & OperationsRedis upgraded to address a reported CVE.FFmpeg 8.0 is now used for modern codec support.Helm compatibility restored with Kubernetes pre‑release versions; charts now pull images from the public repo.Clarified behavior: CVAT_ALLOW_STATIC_CACHE only affects new tasks (existing tasks keep their configured chunking).SDK & CLI Quality of LifeThe SDK now automatically retries certain transient server errors.SDK supports server URLs that include ports (e.g., https://example.com:8443).Performance ImprovementsFaster task creation from the cloud, even when you don’t have a manifest.Lower memory usage when counting objects in tracks during annotation updates and analytics.Manifest Requirements & ClarityFrame width and height are now required in dataset manifests (2D & 3D).Manifests can include an optional original_name field, and error messages at task creation are clearer.Documentation around supported layouts with related images has been improved.Cleanup and DeprecationsYou can no longer upgrade directly from releases prior to v2.0.0—plan a staged path through a 2.x release.Removed legacy, non‑functional API URL signing code.Upcoming change: overly broad filtering of files that merely contain the string related_images is deprecated. CVAT will filter only actual related‑image files according to the input layout.‍FixesExport integrity: Fixed an issue where tracks could leak between jobs on export.Cloud storage workflows: Bulk delete now removes all selected storages; project/task transfers retain the correct storage reference.Related images: Detection is more reliable across all supported layouts for both 2D and 3D media.UI stability & polish: Resolved a crash when loading annotation format metadata.Fixed a sporadic UI error related to reading points.Corrected model card clipping on small displays.Organization description updates now save correctly.Server resilience: The backend starts correctly even when analytics are disabled, and a packaging quirk that produced a misleading pkg_resources error message is handled.Have suggestions or requests for what you'd like to see next? Open an issue on GitHub.

Announcing the new Ultralytics YOLO support for automatic annotation via CVAT agents. Powerful computer vision libraries such as Ultralytics YOLO, Detectron2, and MMDetection have made it easier to train high-performing models for a wide variety of tasks. However, using these models for automated annotation often requires custom code, format conversions, and one-off integrations, especially when labeling workflows span multiple tasks. As a result, many teams fall back on manual labeling because they find automation too complex to adopt at scale. Ultralytics YOLO is one of the most widely used model families in the computer vision community. Until now, CVAT included a single built-in YOLO model for auto-annotation, but expanding beyond that required manual setup. That's why we're excited to announce our new integration with Ultralytics YOLO via the CVAT AI annotation agent. Introducing the new Ultralytics YOLO and CVAT integration With this new integration, you can use native Ultralytics models (YOLOv5, YOLOv8, YOLO11) and third-party YOLO models with Ultralytics compatibility (YOLOv7, YOLOv10, etc.) for automatic image or video annotation for a wide range of computer vision tasks, including: Classification Object detection Instance segmentation Oriented object detection Pose estimation Just pick a YOLO model you want to label your dataset with, connect it to CVAT via the agent, run the agent, and get fully labeled frames or even entire datasets, complete with the right shapes and attributes, and all, in a fraction of the time. Annotation possibilities unlocked This integration opens up multiple workflow optimization and automation opportunities for ML and AI teams. Here are just a few. Pre-label data using the right model for the task Connect the YOLO models that match your annotation goals and run them sequentially to pre-label your data. Each model can be triggered individually through the CVAT interface, allowing you to generate different types of labels for the same dataset without custom scripts or external tools. This works for any YOLO model, out-of-the-box or fine-tuned. Label entire tasks in bulk Working with a large dataset? You don’t have to annotate each frame manually. Apply a YOLO model to the entire task in one step. Just open the Actions menu in your task and select Automatic annotation. CVAT will send the job to the agent and automatically annotate all frames across all jobs in a task, saving you time and reducing repetitive work. Share models across teams and projects Register a model once via a native function and agent, and make it instantly available across your organization in CVAT. Team members can use it in their own tasks without any local setup. Validate model performance on real data Test your fine-tuned YOLO model directly on annotated datasets and compare its predictions side-by-side with human labels in CVAT. Spot mismatches, edge cases, or underperforming classes, all without leaving your annotation environment. How it works Here’s what a typical YOLO auto-annotation setup via agents looks like: Step 1. Write and register the function Start by implementing a native function–a Python script that loads your YOLO model (e.g., yolov8n, yolo11m-seg) and defines how predictions will be generated and returned to CVAT. Then register this function in CVAT using the CLI. Note: You can reuse the same native function both in CLI-based annotation and agent-based mode. Step 2. Start the agent Once the function is registered, launch an agent using the CLI command. This starts a local service that automatically connects to your account in CVAT Online or Enterprise, and listens for annotation requests from CVAT. The agent then runs the model (inside your function), generates predictions, and sends them back to CVAT. For more in-depth information about how to set up automated data annotation with a YOLO or any custom model using a CVAT AI agent, read this article. Step 3. Create or select a task in CVAT Log into your CVAT instance and create a new task (or select an existing one). Upload your images or video, and define the labels you want to annotate (e.g., "person", "car", "helmet"). Depending on your use case, you can define different types of labels such as bounding boxes, polygons, or skeletons to match the expected output from your model. Step 4. Choose the model in the UI Once your task and the job are created and the agent is running, go to the AI Tools panel inside your job. Select the Detector tab and the YOLO model you registered earlier. Step 5. Run AI annotation on selected frames After selecting the model, CVAT sends a request to the running agent. The agent runs the model and returns predictions in the form of shapes (e.g., boxes, polygons, or keypoints), each associated with a label ID. Get started now Ready to speed up your annotation workflow with YOLO? Sign in to your CVAT Online account and try it out yourself. For more information about Ultralytics YOLO models and the tasks they support, check the Ultralytics documentation page. For more information about CVAT AI annotation agents, visit Announcing CVAT AI Agents: A New (and Better) Way to Automate Data Annotation using Your Own Models

TL;DR One-step user off-boarding: remove a user with all their data via a new Django CLI command. Smarter label schema checks: the Raw Labels Editor now blocks invalid configs before they hit the server. Smoother reviews: Issue dialogs auto-reposition to stay fully visible near frame edges. Cleaner APIs: deleted_frames in job meta now only reports frames from the current job segment. Lower backend load: preview requests for Projects/Tasks/Jobs are sent sequentially to avoid spikes. Added Delete a user together with all their resources (admin CLI) Context: Adds a safe server-side flow to remove a user and all associated data in one go. Impact: Admins can fully deprovision accounts without manual cleanup. Built-in validation prevents unsafe deletions (e.g., community users still in orgs with other members; SaaS users with active subscriptions). Example: python manage.py deleteuser <user_id> Changed Raw Labels Editor: stronger validation Context: UI now catches malformed/unsupported label configurations earlier. Impact: Fewer bad payloads reaching the backend; clearer inline feedback for editors. No action needed unless your current config is invalid. Preview requests sent sequentially to reduce server load Context: Fetching previews for Projects/Tasks/Jobs could spike concurrency. Impact: More stable performance under load; slightly lower peak concurrency for preview calls. Fixed Issue dialog never opens off-screen Context: In review mode, dialogs near frame corners could render outside the viewport. Impact: Dialogs now auto-reposition and remain fully visible. No more zooming out to find them. deleted_frames in job meta constrained to the job segment Context: API could return deleted_frames outside the job’s segment in /jobs/{id}/data/meta. Impact: Client code relying on this field now receives only relevant frames. If you worked around the previous behavior, you can simplify your client logic. Notes: Includes a test for video tasks with frame step > 1.

Summer is in full bloom, and we haven’t slowed down. Here’s a quick roundup of our team’s July deliveries across CVAT Online, Enterprise, and Community platforms, and more. CVAT Academy After months of preparation, we’re excited to launch CVAT Academy! CVAT Academy is a hands-on online course to help you and your team master key annotation tools and techniques in CVAT, from your first bounding box to advanced workflows. We’ve published the first module on core CVAT annotation tools like bounding boxes, polygons, and AI tools. You can watch them on the CVAT Academy page or YouTube and leave your likes and comments. Your feedback is valued, and we’ve already made some tweaks based on user input. The best part: after talking to our customers and soon-to-be annotators, we decided to make this course completely free! So, whether you want to improve your annotation skills or onboard new members faster, you can do it at no cost with CVAT Academy. New features (Releases 2.41 + 2.42) SAM2 Object Tracking via AI Agents (CVAT Online) CVAT Online users can now automatically track shapes across video frames with SAM2 or any other tracking model, custom or pre-trained, using AI agents. Read more Improved Navigation on Listing Pages (All Platforms) On listing pages like Tasks, Jobs, Projects, and Models, users can choose how many items to display per page: 10, 20, 50, or 100, and quickly jump to a specific page by entering its number. These updates enhance browsing large datasets and make navigation more efficient. Quick Edit from List Views (All Platforms) We’ve added a dynamic Edit option for Task, Job, and Project list pages. Users can now update key fields directly from the list view without opening each item: Assignee can be changed for tasks, jobs, and projects. State and Stage can be updated for jobs. This simplifies routine updates and reduces extra clicks. Other Changes & Fixes (All Platforms) Enabled multi-threaded image downloading from cloud storage when preparing chunks, enhancing performance. Resolved COCO keypoints export issues with absent keypoints. Updated the organization Actions menu to match the style of other menus. Enabled shortcuts configuration in tag annotation mode. Ensured “Automatically go to the next frame” setting applies correctly when adding the first tag in the tag annotation workspace. Improved performance of GET /api/jobs/<id>/annotations and GET /api/tasks/<id>/annotations with many tracks containing mutable attributes. Enforced email verification when using Basic HTTP authentication, if ACCOUNT_EMAIL_VERIFICATION is set to mandatory, preventing unverified access. Updated Python runtime for the Segment Anything Nuclio function from 3.8 to 3.10 to ensure compatibility and support. API: Deprecated legacy token authentication. Updated API token and session/CSRF token auth schemas for simplified and more secure access. The PATCH and PUT endpoints at /api/tasks/<id>/annotations and /api/jobs/<id>/annotations now enforce that annotation IDs are present when updating and absent when creating. This improves consistency and prevents mismatched operations. Updated API schema for session authentication. To get into all the nitty-gritty details, visit our GitHub changelog.