SAM 2
Meta's second-generation foundation model for general-purpose image and video segmentation. It supports point and box prompts for high-precision object segmentation and tracking.
Timeline (last 90 days)
In the preprocessing stage, the pipeline uses SAM 2 to extract high-resolution vehicle segmentation masks and embeds the vehicle contours into a spatial-conditioning adapter to enforce geometric boundaries at the cross-attention layers.
SAM 3.1 in the community typically refers to community versions that are further fine-tuned or optimized based on the SAM 2 architecture.
In 2024, Meta further released SAM 2, extending it to video segmentation scenarios with improvements in both efficiency and accuracy.
All Facts (3)
In the preprocessing stage, the pipeline uses SAM 2 to extract high-resolution vehicle segmentation masks and embeds the vehicle contours into a spatial-conditioning adapter to enforce geometric boundaries at the cross-attention layers.
50%UnverifiedSAM 3.1 in the community typically refers to community versions that are further fine-tuned or optimized based on the SAM 2 architecture.
50%UnverifiedIn 2024, Meta further released SAM 2, extending it to video segmentation scenarios with improvements in both efficiency and accuracy.
60%