It's too slow.

[kanbekotori]

from mobile_sam import sam_model_registry, SamAutomaticMaskGenerator
import cv2
import time

start_time = time.time()

sam_checkpoint = "./mobile_sam.pt"
mobile_sam = sam_model_registry["vit_t"](checkpoint=sam_checkpoint)
mobile_sam.to(device="cuda")
mobile_sam.eval()
mask_generator = SamAutomaticMaskGenerator(mobile_sam)
image_bgr = cv2.imread("C:\\Users\\kanbe\\Pictures\\aa.jpg")
image_rgb = cv2.cvtColor(image_bgr, cv2.COLOR_BGR2RGB)
masks = mask_generator.generate(image_rgb)

end_time = time.time()
print("{:.2f}s".format(end_time - start_time))

Image file size 54k (1024 x 683).

It took about 7 seconds with the GPU and 70 seconds with the CPU, is this a normal speed?

[zaoyueri]

On my CPU, it will take over a minute to segment an image

[JisuHann]

Have you considered using ONNX model?

[xiangw369]

您是否考虑过使用 ONNX 模型？

能够分享一下相关的代码吗？

[tyronedong]

It takes 10minutes on my cpu for single image

[zaoyueri]

It takes 10minutes on my cpu for single image

large image？

[ghm666]

On my GPU, it take 1.6s to segment an image， img_path = "./notebooks/images/picture1.jpg"