Segment Anything
The Segment Anything project aims to democratize image segmentation in computer vision, a core task used across various applications such as scientific imagery analysis and photo editing. Traditionally, accurate segmentation models require specialized expertise, AI training infrastructure, and large amounts of annotated data. This project introduces a new task, dataset, and model for image segmentation to overcome these challenges and make segmentation more accessible.
The researchers are releasing the Segment Anything Model (SAM) and the Segment Anything 1-Billion mask dataset (SA-1B), the largest segmentation dataset to date. These resources will enable a wide range of applications and further research into foundational models for computer vision. The SA-1B dataset is available for research purposes, while the SAM is provided under the permissive Apache 2.0 open license. Users can explore the demo to try SAM with their own images.
Paper link: https://arxiv.org/abs/2304.02643
Code link: https://github.com/facebookresearch/segment-anything
Demo link: https://segment-anything.com/demo
Blogpost link: https://ai.facebook.com/blog/segment-anything-foundation-model-image-segmentation/
Dataset link: https://ai.facebook.com/datasets/segment-anything/
A detailed unofficial overview of the paper: https://andlukyane.com/blog/paper-review-sam
#deeplearning #cv #pytorch #imagesegmentation #dataset
The Segment Anything project aims to democratize image segmentation in computer vision, a core task used across various applications such as scientific imagery analysis and photo editing. Traditionally, accurate segmentation models require specialized expertise, AI training infrastructure, and large amounts of annotated data. This project introduces a new task, dataset, and model for image segmentation to overcome these challenges and make segmentation more accessible.
The researchers are releasing the Segment Anything Model (SAM) and the Segment Anything 1-Billion mask dataset (SA-1B), the largest segmentation dataset to date. These resources will enable a wide range of applications and further research into foundational models for computer vision. The SA-1B dataset is available for research purposes, while the SAM is provided under the permissive Apache 2.0 open license. Users can explore the demo to try SAM with their own images.
Paper link: https://arxiv.org/abs/2304.02643
Code link: https://github.com/facebookresearch/segment-anything
Demo link: https://segment-anything.com/demo
Blogpost link: https://ai.facebook.com/blog/segment-anything-foundation-model-image-segmentation/
Dataset link: https://ai.facebook.com/datasets/segment-anything/
A detailed unofficial overview of the paper: https://andlukyane.com/blog/paper-review-sam
#deeplearning #cv #pytorch #imagesegmentation #dataset
🔥13👍5❤1
Forwarded from Machinelearning
NVIDIA представила новый подход к обучению моделей для сложных математических задач, заняв первое место в конкурсе Kaggle AIMO-2.
Секрет — в огромном датасете OpenMathReasoning, который состоит из 540 тыс. уникальных задач с Art of Problem Solving, 3,2 млн. многошаговых решений (CoT) и 1,7 млн. примеров с интеграцией кода (TIR).
Для сравнения: это в разы больше, чем в популярных аналогах MATH и GSM8K. Все это дополнено 566 тыс. примеров для обучения генеративному выбору решений (GenSelect) — методу, который лучше, чем классическое голосование большинством.
OpenMathReasoning создавался тщательно и ответственно. Сначала задачи фильтровались через Qwen2.5-32B, чтобы убрать простые или дублирующие бенчмарки. Затем DeepSeek-R1 и QwQ-32B генерировали решения, а итеративная тренировка с жесткой фильтрацией улучшала качество. Например, код в TIR-решениях должен был не просто проверять шаги, а давать принципиально новые вычисления — вроде перебора вариантов или численного решения уравнений.
Модели OpenMath-Nemotron (1,5B–32B параметров), обученные на этом наборе данных показали SOTA-результаты. 14B-версия в режиме TIR решает 76,3% задач AIME24 против 65,8% у базового DeepSeek-R1. А с GenSelect, который анализирует 16 кандидатов за раз, точность взлетает до 90%. Даже 1,5B-модель с GenSelect обгоняет 32B-гиганты в отдельных тестах.
@ai_machinelearning_big_data
#AI #ML #Math #Dataset #NVIDIA
Please open Telegram to view this post
VIEW IN TELEGRAM
Please open Telegram to view this post
VIEW IN TELEGRAM
👍5🔥3❤1