Загрузка видео...

Не удалось загрузить видео

На главную

we are introducing a state-of-the-art real-time object detection model, RF-DETR RF-DETR outperforms all existing object detection models on real world datasets and is the first real-time model to achieve 60+ Average Precision on COCO talked w/ NVIDIA about it at GTC:

18,231 просмотров • 1 год назад •via X (Twitter)

Комментарии: 11

Фото профиля Joseph Nelson
Joseph Nelson1 год назад

RF-DETR builds on the work of real-time detection transformers like RT-DETR and LW-DETR. we show that large scale pretraining from a DINOv2 backbone and smaller input resolution sizes produces higher results at real-time speeds comparable with YOLO models (inclusive of NMS).

Фото профиля Joseph Nelson
Joseph Nelson1 год назад

we also introduce a new benchmark COCO is becoming saturated and doesn't measure performance across domains. RF100-VL is sampled from 500k open source vision datasets representative of how vision is being applied: aerial, labs, wildlife, . . . RF100-VL is Domain Adaptability

Фото профиля Joseph Nelson
Joseph Nelson1 год назад

RF-DETR is comparable to SOTA realtime transformers on COCO (and SOTA at largest size with 720p input resolution), which are ahead of YOLO models, while adapting to new domains the best of any model. we show RF-DETR is the best real-time object detection model on real world data

Фото профиля Joseph Nelson
Joseph Nelson1 год назад

we expect (and hope!) models will build on and pass RF-DETR's SOTA result. the training code and RF100-VL evals are open source Apache 2.0 to do so. at roboflow, we believe more SOTA open source models is good for the ecosystem + users. we will continue to support any/all models

Фото профиля Joseph Nelson
Joseph Nelson1 год назад

RF-DETR is available for training and deployment in roboflow's hosted platform, via Colab Notebook, and on GitHub. give us a star, it's not everyday a tiny ML team passes prestigious Chinese research labs :) ↳ ↳

Фото профиля Lucid Scientific, Inc.
Lucid Scientific, Inc.1 год назад

Expand the possibilities of your metabolic research. Resipher tracks real-time cellular oxygen consumption in standard 96-well plates, delivering continuous real-time data directly from your incubator. Request a free virtual demo or quote today >>

Фото профиля Alex Volkov (Thursd/AI)
Alex Volkov (Thursd/AI)1 год назад

Congrats on the release @roboflow team! Thanks for joining us today @josephofiowa The full segment is live on @thursdai_pod

Фото профиля Joseph Nelson
Joseph Nelson1 год назад

@roboflow @thursdai_pod great ThursdAI, as always!

Фото профиля Calvin Chen
Calvin Chen1 год назад

wow who are these guys and why are they so cool

Фото профиля Turner Novak 🍌🧢
Turner Novak 🍌🧢1 год назад

Sounds awesome, will have to check this out

Фото профиля Joseph Nelson
Joseph Nelson1 год назад

it should handle your needs perfectly

Похожие видео