MedSAM inference

MedSAM inference will be done here

from datasets import load_dataset

dataset = load_dataset("nielsr/breast-cancer", split="train")

idx = 10

image = dataset[idx]["image"]
label = dataset[idx]["label"]
msk = np.array(label)
an_img = overlay_mask_border_on_image_frm_img(
    image, msk,
)
show_(an_img)
#msk.shape, image.size

cntrs = find_contours_binary(msk.astype(np.uint8))[0]
x, y, w, h = frm_cntr_to_bbox(cntrs)

def get_model():
    model = SamModel.from_pretrained("wanglab/medsam-vit-base")
    return model

---

source

get_bounding_box

 get_bounding_box (ground_truth_map:numpy.ndarray)

Get bounding box from mask

---

source

get_prediction

 get_prediction (model:transformers.models.sam.modeling_sam.SamModel,
                 model_name:str, image:PIL.Image.Image, boxes:List[int],
                 device:Optional[str]=None, threshold:float=0.5)

	Type	Default	Details
model	SamModel
model_name	str		checkpoint in hugggingface
image	Image
boxes	List
device	Optional	None
threshold	float	0.5
Returns	Tuple

model = get_model()

preds_ = get_prediction(
    model=model,
    model_name='wanglab/medsam-vit-base',
    image=image,
    boxes=[get_bounding_box(msk)],
    device='cpu',
    threshold=0.9

)

an_img = overlay_mask_border_on_image_frm_img(image, preds_)
show_(an_img)