SAM21ModelSize
Enums · AiDotNet.Enums
Defines the backbone size variants for SAM 2.1 (Segment Anything Model 2.1).
For Beginners
SAM 2.1 is an updated version of SAM 2 with refined training recipes that improve segmentation accuracy. It uses the same Hiera backbone architecture as SAM 2 but with better-tuned checkpoints.
How It Works
Technical Details: Same Hiera backbone as SAM 2 with improved training procedures. Supports both image and video segmentation with memory attention for temporal consistency.
Reference: Ravi et al., “SAM 2: Segment Anything in Images and Videos”, Meta AI, 2024.
Fields
| Field | Summary |
|---|---|
SAM21ModelSize BasePlus |
Base Plus variant (81M params). |
SAM21ModelSize Large |
Large variant (224M params). |
SAM21ModelSize Small |
Small variant (46M params). |
SAM21ModelSize Tiny |
Tiny variant (39M params). |
int value__ |