SAM
A foundation model for image segmentation released by Meta AI in 2023. SAM supports zero-shot segmentation of arbitrary objects using diverse prompt types, including points and bounding boxes. It represents a paradigm shift toward promptable, general-purpose vision foundation models and was trained on a large-scale dataset of over one billion masks.
Core Facts
SAM (Segment Anything Model) is a foundation model released in 2023 capable of zero-shot segmentation of any object in an image; SAM 2, released in 2024, supports video segmentation.
SAM (Segment Anything Model) is a model open-sourced by Meta that achieves zero-shot segmentation of arbitrary objects through large-scale pre-training on 11 million images and 1 billion masks.
Meta's research division developed models such as SAM (Segment Anything Model) and ImageBind.
Timeline (last 90 days)
Meta open-sourced the vision models DINOv2 and SAM.
Converting segmentation masks to detection boxes with SAM usually takes the mask's minimum bounding rectangle, but this may not be precise enough for overlapping objects or non-rectangular objects.
QAT regularizes by explicitly introducing quantization noise, forcing the model to converge to flatter regions, which is akin to the idea of Sharpness-Aware Minimization (SAM).
RADIO is a visual foundation model proposed by NVIDIA that generates visual representations balancing semantic alignment and fine-grained features by distilling knowledge from multiple vision-language pretrained models such as CLIP, DINOv2, and SAM.
SAM has zero-shot generalization capability and can generate segmentation masks for any object in an image via clicks, bounding boxes, or text prompts, without requiring fine-tuning for specific categories.
In AI generation workflows, the SAM series is often used as a preprocessor to first extract precise boundaries of target regions from reference images before passing them to subsequent generation or editing steps.
SAM was trained on a dataset of more than 1 billion masks.
SAM is a general-purpose image segmentation model released by Meta AI in April 2023, based on a Vision Transformer architecture with 1.1 billion parameters.
SAM itself only addresses the question of which pixels belong to the target and does not provide any 3D geometric information; it must be used in conjunction with subsequent calibration and pose estimation modules.
The SAM series is a general-purpose image segmentation model released by Meta; its core concept is to segment everything, capable of segmenting any object in an image without training on specific categories.
1 more timeline events
All Facts (20)
SAM (Segment Anything Model) is a foundation model released in 2023 capable of zero-shot segmentation of any object in an image; SAM 2, released in 2024, supports video segmentation.
65%VerifiedSAM (Segment Anything Model) is a model open-sourced by Meta that achieves zero-shot segmentation of arbitrary objects through large-scale pre-training on 11 million images and 1 billion masks.
65%VerifiedMeta's research division developed models such as SAM (Segment Anything Model) and ImageBind.
65%VerifiedSAM (Segment Anything Model) is a semantic segmentation model developed by Meta.
65%VerifiedSAM3 is the third-generation product of the SAM series, with its core breakthrough being the extension of segmentation capability from static images to dynamic video and the introduction of natural language understanding
65%UnverifiedSegment Anything（SAM）是Meta推出的通用图像分割基础模型，能够零样本分割任意物体
90%UnverifiedFastSAM是Meta SAM模型的轻量化版本，能够在毫秒级时间内完成图像中任意目标的分割掩码生成
85%Unverified工作流中的抠图节点依赖语义分割模型（如SAM、U²-Net等），能够自动识别图像中的主体对象并去除背景，输出带透明通道的PNG图像
80%UnverifiedThe SAM series is a general-purpose image segmentation model released by Meta; its core concept is to segment everything, capable of segmenting any object in an image without training on specific categories.
60%UnverifiedSAM's architecture consists of three components: a heavyweight image encoder based on Vision Transformer, a flexible prompt encoder, and a lightweight mask decoder.
60%UnverifiedThe original SAM version released by Meta in 2023 was trained on the SA-1B dataset of over 1 billion masks
60%UnverifiedMeta open-sourced the vision models DINOv2 and SAM.
50%UnverifiedConverting segmentation masks to detection boxes with SAM usually takes the mask's minimum bounding rectangle, but this may not be precise enough for overlapping objects or non-rectangular objects.
50%UnverifiedQAT regularizes by explicitly introducing quantization noise, forcing the model to converge to flatter regions, which is akin to the idea of Sharpness-Aware Minimization (SAM).
50%UnverifiedRADIO is a visual foundation model proposed by NVIDIA that generates visual representations balancing semantic alignment and fine-grained features by distilling knowledge from multiple vision-language pretrained models such as CLIP, DINOv2, and SAM.
50%UnverifiedIn AI generation workflows, the SAM series is often used as a preprocessor to first extract precise boundaries of target regions from reference images before passing them to subsequent generation or editing steps.
50%UnverifiedSAM has zero-shot generalization capability and can generate segmentation masks for any object in an image via clicks, bounding boxes, or text prompts, without requiring fine-tuning for specific categories.
50%UnverifiedSAM was trained on a dataset of more than 1 billion masks.
50%UnverifiedSAM itself only addresses the question of which pixels belong to the target and does not provide any 3D geometric information; it must be used in conjunction with subsequent calibration and pose estimation modules.
50%UnverifiedSAM is a general-purpose image segmentation model released by Meta AI in April 2023, based on a Vision Transformer architecture with 1.1 billion parameters.
50%