Research Scientist

Launch: RF-DETR Keypoint in Roboflow. RF-DETR Keypoint is a real-time, end-to-end pose model that beats YOLO26-pose on both accuracy and speed, learns calibrated per-keypoint uncertainty from your data, and ships under Apache 2.0. You can label, train, and deploy it in Roboflow today. Today Roboflow, Inc. is launching RF-DETR Keypoint support in Roboflow. RF-DETR Keypoint is a real-time, end-to-end keypoint detection model that extends the RF-DETR family from boxes and masks. You can use RF-DETR Keypoints across the Roboflow platform: label your skeletons in Annotate, train on your own data with Train, and deploy with Inference and Workflows. RF-DETR Keypoint is releasing as a preview, the same way its instance segmentation model did, because Roboflow, Inc. want real-world usage to shape the final family of checkpoints. You can start building with it now. For the architecture, the self-calibrating loss, and the full benchmark methodology, read the RF-DETR Keypoint technical deep dive. Label, Train, and deploy in Roboflow. RF-DETR Keypoint runs through the full Roboflow workflow: Label your skeletons. Define the keypoints and the skeleton for your object in Annotate, whether that is a 17-point person or 20 points on a basketball court. There is no requirement to match COCO. Train on your data. Fine-tuning RF-DETR Keypoint works the same way as it does for detection and segmentation. Deploy where you run. Serve the model with Inference on the cloud or the edge, and chain it with logic and other models in Workflows to turn poses into applications: rep counting, ergonomic checks, sports analytics, robot guidance, gauge reading. What is RF-DETR Keypoint? RF-DETR Keypoint builds a keypoint head directly into the detection transformer. For every object it detects, it predicts a structured set of keypoints in a single forward pass: no NMS, no heatmaps, no post-hoc grouping of points into instances. Detection and pose are learned jointly, so the keypoints even feed back to make the detections better. It is not limited to people. The COCO checkpoint covers the 17-keypoint human skeleton because that is the benchmark everyone can check, but the architecture supports arbitrary skeletons: any number of keypoints, on any object class, with multiple keypointed classes in one model. License-plate corners, surgical instrument tips, robot-arm joints, and gauge needles are the point. How RF-DETR Keypoint is different. Three things set RF-DETR Keypoint apart from other pose models. It beats the newest YOLO pose models on accuracy and speed. On COCO Keypoints, measured end-to-end on an NVIDIA T4 with TensorRT FP16, the preview checkpoint at 576x576 reaches 71.8 AP at 9.8 ms, ahead of YOLO26x-pose (the largest model in the newest YOLO pose family) at 71.0 AP and 10.6 ms. Scaled up to 888x888 it reaches 74.2 AP, within 0.6 AP of the state-of-the-art academic model GroupPose Swin-L while running roughly 13 times faster. It learns its own keypoint uncertainty, and hands it to you. Most pose models depend on per-keypoint tolerance constants that were hand-measured on COCO and quietly fall back to a uniform guess the moment your skeleton is not COCO's. RF-DETR Keypoint instead predicts a full distribution for each keypoint and calibrates that spread from your data, so the loss that produces the benchmark numbers is the same loss you fine-tune with. Each keypoint comes back with a confidence ellipse and a usable 2D covariance, a drop-in observation model for a tracker, a Kalman filter, or a skeleton fit, plus separate signals for whether a keypoint is findable at all versus merely occluded. One checkpoint runs at every speed. Like the rest of the RF-DETR family, it is trained with weight-sharing neural architecture search, so a single set of weights runs across resolutions from about 4.5 ms to 26 ms with no retraining. Train once, get a family of options. RF-DETR is Apache 2.0: ship it anywhere. RF-DETR Keypoint Preview is released under the Apache 2.0 license, code and weights, free for commercial use, with no copyleft obligations and deployable inside closed-source products. The YOLO family are AGPL-3.0, which in practice means open-sourcing the application you build around the model or buying an enterprise license, even for many internal commercial uses. If licensing has been the thing standing between your team and shipping a pose model, it is not anymore. Conclusion. RF-DETR Keypoint is launching as a preview. The architecture and training recipe are performing well in its evaluations, but real-world usage surfaces things benchmarks do not, and that feedback will shape what comes next: checkpoints across the full accuracy-latency curve, NAS-enabled training on the platform, and purpose-built models for popular keypoint datasets. Cite this post. Use the following entry to cite this post in your research: Contributing Writer. (Jun 22, 2026). Launch: RF-DETR Keypoint in Roboflow. Roboflow Blog: https://blog.roboflow.com/launch-rf-detr-keypoint-in-roboflow/ Stay Connected Get the Latest in Computer Vision First Contributing Writer Topics.

Launch: Use YOLO26 Semantic Segmentation with Roboflow. You can now label data for semantic segmentation, train YOLO26 semantic segmentation models, and deploy them with Roboflow. Roboflow, Inc. is excited to announce support for YOLO26 semantic segmentation in Roboflow. You can now label data for semantic segmentation, train YOLO26 semantic segmentation models, and deploy them with Roboflow, all in one place. What is YOLO26 Semantic Segmentation? Semantic segmentation assigns a class label to every pixel in an image, producing a dense class map that covers the entire scene. Rather than drawing a box around an object or outlining each individual instance, a semantic segmentation model classifies every pixel into a category, grouping all pixels of the same class together regardless of how many distinct objects are present. This differs from instance segmentation, which separates individual objects of the same class into distinct masks. Because semantic segmentation produces a single dense class map for the full image, it is well suited to scene-level understanding tasks such as autonomous driving, land-cover mapping, and medical imaging, where the goal is to understand every region of an image rather than to count or track separate objects. YOLO26 is the latest model architecture from the team behind YOLO11 and YOLOv8. Its semantic segmentation models extend the YOLO family into pixel-wise, dense prediction, bringing the real-time performance the architecture is known for to whole-scene labeling. Label data for YOLO26 Semantic Segmentation with Roboflow. Building an accurate semantic segmentation model starts with high-quality labeled data. Roboflow provides an extensive suite of annotation tools to help teams label images for semantic segmentation, including AI-assisted and SAM-powered labeling to accelerate the process. Existing datasets can also be brought in and converted, and trained models can be used as label assistants to speed up labeling further. Centralizing annotation in Roboflow gives teams a consistent place to manage datasets, review labels, and prepare data for training without stitching together separate tools. Train YOLO26 Semantic Segmentation models with Roboflow. Once your dataset is ready, you can train YOLO26 semantic segmentation models on Roboflow's hosted training platform. Roboflow manages the underlying training infrastructure and GPU access, so teams can train models without provisioning or maintaining their own hardware. When training is complete, the model is available for testing directly in the Roboflow web interface, making it straightforward to evaluate performance before moving toward deployment. Deploy YOLO26 Semantic Segmentation models with Roboflow. After a model has trained, it can be deployed through the Roboflow cloud API or run on your own hardware with Roboflow Inference. Inference supports deployment across CPU and GPU devices, including edge hardware, so models can run close to where images are captured. For the lowest latency, Roboflow recommends deploying on device with Inference. YOLO26 semantic segmentation models can also be incorporated into Roboflow Workflows, a low-code interface for building multi-step computer vision pipelines and applications that can then be deployed in the cloud or on your own hardware. Get started with YOLO26 Semantic Segmentation. With support for YOLO26 semantic segmentation, Roboflow covers the full path from labeling data to training a model to deploying it in production. Semantic segmentation brings dense, pixel-level scene understanding to the YOLO family, and Roboflow makes each step of working with these models available in a single platform. Cite this post. Use the following entry to cite this post in your research: Contributing Writer. (Jun 11, 2026). Launch: Use YOLO26 Semantic Segmentation with Roboflow. Roboflow Blog: https://blog.roboflow.com/use-yolo26-semantic-segmentation-with-roboflow/ Stay Connected Get the Latest in Computer Vision First Contributing Writer Topics.

Vision Events: Turning Images into Insights. Today Roboflow, Inc. is announcing Roboflow Vision Events, a centralized view into what's happening across your vision deployments. By storing model predictions, images, and relevant metadata in one place, Vision Events transforms the output of your computer vision applications into searchable, historical business insights that drive better operational decisions. A secure, searchable record of your vision deployments. When deploying a computer vision application, you need a reliable place to store model outputs and connect them to your broader goals. For example, if your vision application monitors manufacturing processes across various facilities, your team might want to answer operational questions like, "How many quality issues did our vision app catch today?", "Which facility experienced the most downtime this month?", or "How can we verify compliance issues detected on a specific assembly line?" Traditionally, answering those questions meant building a custom solution to store model predictions alongside the associated image files and relevant metadata. Standard tools often make it challenging to co-locate tabular data with the actual production images from your vision model. That is why Roboflow, Inc. built Vision Events. Now you can aggregate data from all your vision applications running across different cameras or geographic regions into a single, centralized hub. It gives you one place to store model predictions, associated input and output images, device info, and custom metadata. Vision Events closes the loop, connecting the initial problem you set out to solve directly to high-level operational visibility. See all events, images, and metadata in one place. If you have similar data coming from multiple deployments, Vision Events lets you group that information into specific views. These views consolidate the results of your vision model, related images, and additional contextual details. You can easily track standard information like the date, time, and device that captured the imagery, as well as customizable properties like shift times, lot numbers, or manufacturing lines. Filter and find specific events. If you need to review specific information, you can filter your data by a wide range of properties, view the results, and save those filters for easy access later. This makes it easy to not only review the overall performance of your model, but also pinpoint specific events when necessary. For example, if you are tracking quality issues in a battery manufacturing facility and need to review images from a specific time, assembly line, or range of serial numbers, you can filter by the relevant metadata to quickly locate the exact information you need. While reviewing data in Vision Events, if you notice false positives or negatives, you can select images and add them directly to a training dataset. This allows you to simplify the process of continuously improving your vision model. How to bring data into Vision Events. Depending on how your vision application is deployed, there are several options for feeding data into the system. * Workflows: For applications built in the drag-and-drop Workflows editor, you can use the Vision Events Workflow block. * Developer Tools: For more customizable integrations, you can use the REST API and Python SDK to bring data into Vision Events. * Edge Device Backup: For applications running on edge hardware, you can automatically synchronize events from each device using the Edge Device Backup option. Start turning images into insights today. Vision Events is available to all Roboflow users. If you want to get started, log into Roboflow and head to the Vision Events tab. Additional information is available in documentation and pricing. To discuss customizable enterprise deployments, feel free to contact Roboflow, Inc.. Cite this post. Use the following entry to cite this post in your research: Patrick Deschere. (Apr 20, 2026). Vision Events: Turning Images into Insights. Roboflow Blog: https://blog.roboflow.com/vision-events/ Stay Connected Get the Latest in Computer Vision First Patrick Deschere Patrick makes content about solving business challenges with vision AI. He spends his time hosting webinars, editing slides, and drawing bounding boxes around objects. Topics.