AI_SAM_test - 100-hours-a-week/16-Hot6-wiki GitHub Wiki
μμ λ¬Έμλ‘ μ΄λ : AI Wiki
Comparison of Segmentation Anything 2.1 Model Weights
-
Due to VRAM limitations, resizing the input image to 1024x1024 is required
-
SAM 2.1 provides multiple model weight options based on different backbone architectures
-
Original Image resized to 1024x1024
Comparison Table
Large | Base_Plus | Small | Tiny | Large_generation option | |
---|---|---|---|---|---|
Image | |||||
VRAM | About 6.1GB | About 4.2GB | About 4.4GB | About 4.3GB | About 14.4GB |
- While the generation time was not explicitly measured, inference on an NVIDIA A100 GPU takes approximately 4 seconds per image.