Roboflow creates software-as-a-service products to make building with computer vision easy. Over 1,000,000 developers use Roboflow to manage image data, annotate and label datasets, apply preprocessing and augmentations, convert annotation file formats, train a computer vision model in one-click, and deploy models via API or to the edge. https://roboflow.com
External link for Roboflow
or
Remote, US
Roboflow reposted this
Build, test, and deploy vision apps... by chatting with a coding agent. The new Roboflow MCP Server connects Claude, Codex, Cursor (and other assistants that work with MCP) to Roboflow. That way, your preferred coding agent can create projects, label data, train custom models, build workflows, and deploy to the cloud or edge via a conversation. Blog: https://lnkd.in/gu4M8YyS Docs: mcp.roboflow.com
RF-DETR is now in transformers Another milestone on the journey to bring SOTA and truly open source vision models to everyone. Having it land natively in transformers makes it that much easier to reach for wherever you're building. This is an important step in making the next wave of AI more than just text on a screen. Getting there takes infrastructure that's open by default and models that run in real-time so that software can see, understand, and act in the physical world. More RF-DETR announcements and releases very very soon.
RF-DETR just landed to Hugging Face transformers 🔥 sota real-time detection & segmentation models by Roboflow 💜 to celebrate this, we shipped real-time webcam streaming demo and fine-tuning tutorials on satellite imagery segmentation and mobile UI detection 🙌🏻 > play with our real-time demo > fine-tune the models on your use case with our tutorials (takes a toaster's VRAM) > or just hand them to your agents 😄 tutorials → https://lnkd.in/eCHKec-8 models and the demo → https://lnkd.in/e82MsJwv docs → https://lnkd.in/edvKRrz8
Roboflow reposted this
New video: Track objects in video with SORT and OC-SORT If you are counting or tracking the movement of objects in video feeds, picking the right tracking algorithm (and tuning the settings) is crucial. In this tutorial, Machine Learning Engineer Lee Clement of Roboflow talks about choosing between open-source algorithms ByteTrack, SORT, and OC-SORT and then shows how to test them in the cloud using Roboflow Workflows. See the full video here: https://lnkd.in/eRpjqqZi
Roboflow reposted this
NYC ✅ Roboflow Summit ✅ great craic ✅ 🗽 I spoke last week at Roboflow Summit - a fireside chat with Joseph Nelson. Really fun conversation and an even better event. We’ve been using Roboflow for years at FloVision, and it’s been a real accelerator for our ML pipeline: • faster annotation + iteration • smoother collaboration • less friction turning edge cases into better models Big thanks to Joseph + the whole Roboflow team for having me - you’ve built a great product, and this summit showed us what it looks like to operate at scale. Bar set. 🚀 FloVision Solutions 🤝 Roboflow (more to come)
Roboflow reposted this
Using vision-language models (like Qwen) to navigate and interact with ui elements. In the quick demo below, Machine Learning Engineer Matvei Popov from Roboflow walks through a project where he built a visual AI agent that takes over his mouse to kick off a model training job. Here is how it works: ⛓️ A Qwen 3.5 pipeline created in Roboflow Workflows. ⚙️ Local inference running on a GPU machine via Roboflow Inference. 📝 A custom script that prompts the model with specific instructions. This is a fun little demo showing how VLMs can be used not only to automate repetitive digital tasks but also to power real-world applications, like enabling robots to interact with physical machinery. Full breakdown here: https://lnkd.in/dprSBdhc
Roboflow reposted this
This week's webinar: Roboflow MCP + Claude Building computer vision applications requires a lot of manual work before you get to your first prediction: extracting frames, annotating, training, and writing deployment code. Roboflow Engineer Tony França will demonstrate how to offload these tasks using the new Roboflow MCP Server. You'll see how to connect Claude to your Roboflow workspace to execute commands like projects_create, images_upload, and models_train. Join the live session to watch an end-to-end demo taking a raw video file to a deployed model using an AI coding agent. Register here: https://luma.com/mt2k9yld
Roboflow reposted this
Trackers 2.4.0 at Roboflow are released! 🔥 🆕 Among the others, we bring you BoT-SORT tracker with camera motion compensation and on-demand hyperparameter search. 🎥 Thanks to its camera motion compensation (training-free), BoT-SORT can handle the camera movements in your videos, adjust the predicted trajectories and perform more accurate associations with detected objects. ⚙️ With hyperparameter search, you can now easily tune the tracker of your choice on your dataset to further improve its performance. It is a pure pleasure to work with such an amazing team and make these things happen! Many thanks to Alexander Bodner, Omkar Kabde, Christoph Deil, Piotr Skalski and Jiri Borovec for great collaboration on this release! Links, more details and other cool stuff in the comments below ⏬ 🔍 And how about the re-ID in BoT-SORT? While camera motion compensation is already a solid enhancement, we observed that the usage of re-ID in the multi-object tracking algorithms can be improved and as such it will come during one of the next releases, thus stay tuned! #tracking #multiobjecttracking #botsort #cmc #camera #motion #compensation #roboflow #hyperparameter #tuning #search #sportsmot
Roboflow reposted this
Google Earth is so underrated and it freaks me out. Week 5 of the challenge and I've been playing around looking at random forests, factory sides, urban sprawls, and historic monuments. There's so much unused imagery and data on every street, block, and jungle in the world, just waiting to be extracted. I'm also a big fan of Meta's SAM3, so I thought to myself: why not merge the two? GeoSeg takes in KML files, reads your natural language description of your area of interest, and automatically segments it for you. It also estimates total area vs % coverage. Meaning you can analyze any object you find on Google Earth. This one might be too niche but I had hella fun doing it. How I did it: - Google Earth Community for guidance on file types - Roboflow to host my SAM3 inference HTTP - MyGeoData for KML to coordinates - Gemini for the language processing piece - Lovable for the UI (might be a cliche at this point) 4 more weeks to go, we roll
Roboflow reposted this
Live session this week: How to track objects in video with SORT and OC-SORT algorithms, presented by Lee Clement, Machine Learning Engineer at Roboflow. Come hang out with us if you're interested in counting or tracking the movement of specific objects in video feeds. Event link: https://luma.com/pd4hodix
We are working on something new and Asfandiyar Khan built this application in 10 minutes. It used to take ~10 days to deliver a car counting project from scratch (data acquisition, labeling, training, application logic).
Wanted to validate a quick idea so I prompted our new Agent and it built this for me from scratch in 10 minutes (left and right lanes are marked from driver's pov). Three months ago, if you told me I’d be able to do this so quickly, I’d politely call you a liar. A lot of production-deployed CV work has always been plumbing between pieces. We're now at a point where that layer gets easier, and teams get more room to focus on the parts that determine whether a model performs exceptionally well in production: data quality, training configs, and deployment. more coming soon on this front ... stay tuned!