metadata
license: apache-2.0
pipeline_tag: mask-generation
tags:
- sam2
SAM2-Hiera-tiny
This repository contains tiny variant of SAM2 model. SAM2 is the state-of-the-art mask generation model released by Meta.
Usage
You can use it like below. First install packaged version of SAM2.
pip install samv2 huggingface_hub
Each model requires different classes to infer.
from huggingface_hub import hf_hub_download
from sam2.build_sam import build_sam2
from sam2.sam2_image_predictor import SAM2ImagePredictor
hf_hub_download(repo_id = "merve/sam2-hiera-tiny", filename="sam2_hiera_tiny.pt", local_dir = "./")
ckpt = f"./sam2_hiera_tiny.pt"
config = "sam2_hiera_t.yaml"
sam2_model = build_sam2(config, ckpt, device="cuda", apply_postprocessing=False)
predictor = SAM2ImagePredictor(sam2_model)
# it accepts coco format
box = [x1, y1, w, h]
predictor.set_image(image)
masks = predictor.predict(box=box,
multimask_output=False)
For automatic mask generation:
from huggingface_hub import hf_hub_download
from sam2.build_sam import build_sam2
from sam2.automatic_mask_generator import SAM2AutomaticMaskGenerator
hf_hub_download(repo_id = "merve/sam2-hiera-tiny", filename="sam2_hiera_tiny.pt", local_dir = "./")
sam2_checkpoint = "../checkpoints/sam2_hiera_tiny.pt"
model_cfg = "sam2_hiera_t.yaml"
sam2 = build_sam2(model_cfg, sam2_checkpoint, device ='cuda', apply_postprocessing=False)
mask_generator = SAM2AutomaticMaskGenerator(sam2)
masks = mask_generator.generate(image)
Resources
The team behind SAM2 made example notebooks for all tasks.
See image predictor example for full example on prompting.
See automatic mask generation example for generating all masks.