We publish original state-of-the-art research, have invested $1M+ into researchers and academics, and are cited by leading global institutions.
In early 2025, we released RF-DETR, a state-of-the-art real-time object detection model. The first to achieve real-time 60+ mAP on COCO. Open-source, Apache 2.0 license.
Deploy Today
Yann LeCun
Chief AI Scientist at Meta
RF-DETR: DINOv2 pretrained encoder + multiscale DETR architecture = SOTA on fast object detection. @nicolas.carion
Nick Urena
Machine Learning Engineer @ USG
We benchmarked RF-DETR vs the YOLO models we were using for our industrial use cases. We're migrating to RF-DETR in every place we can given seeing its superior performance.
Merve Noyan
Open-sourceress at Hugging Face 🤗
do not sleep on RF-DETR 🔥 it's a sota, real-time and most importantly, open-source object detector released by @roboflow @skalskip92. my vibe tests on very noisy images and videos have passed 👏 currently being integrated to transformers 🤗
Evaluating vision models on comprehensive real-world datasets.
A new object detection benchmark that measures model domain adaptability to real world problems. RF100-VL was designed by researchers from Roboflow and Carnegie Mellon University.
Vision AI Checkup measures how well new multimodal models perform at real world use cases. Our assessment consists of dozens of images, questions, and answers that we benchmark against models and is meant as an unofficial "vibe test" of VLM performance.
Advancing the state-of-the-art in object recognition with a new way to benchmark computer vision models across domains and task targets.
Upload an image and enter classes to compare how AI vision models compare to one another.
Our leaderboard shows the accuracy of top-performing models on the industry-standard Microsoft COCO object detection benchmark.
Every day, researchers and academics are using Roboflow to advance their work.
We have invested over $1M into new research leveraging Roboflow.
Our team presents research and industry insights at leading computer vision conferences.
RF-DETR will be presented at ICLR 2026, showcasing neural architecture search for real-time detection transformers.
RF100-VL was presented at NeurIPS 2025, introducing a multi-domain object detection benchmark for vision-language models.
Presented at the Vision Workshop and RF100-VL was showcased as a non-archival poster at the 4th Multimodal Foundation Models Workshop.
Roboflow and Carnegie Mellon University are collaborating to release the second iteration of the Foundational Few-Shot Object Detection Challenge at CVPR 2025.
Peter Robicheaux and Isaac Robinson of Roboflow and Vik Korrapati at Moondream recap the best work of 2024 in frontier/open model vision work.
