diff --git a/demo/video_matting/webcam.py b/demo/video_matting/webcam.py
index fc09aeb..1bf5438 100644
--- a/demo/video_matting/webcam.py
+++ b/demo/video_matting/webcam.py
@@ -1,4 +1,6 @@
+
 import cv2
+import argparse
 import numpy as np
 from PIL import Image
 
@@ -8,6 +10,11 @@ import torchvision.transforms as transforms
 
 from src.models.modnet import MODNet
 
+parser = argparse.ArgumentParser()
+parser.add_argument('--ckpt-path', type=str, default="./pretrained/modnet_webcam_portrait_matting.ckpt", help='path of pre-trained MODNet')
+parser.add_argument('--cpu', action='store_true', default=False, help="use cpu inferece")
+args = parser.parse_args()
+
 
 torch_transforms = transforms.Compose(
     [
@@ -17,10 +24,12 @@ torch_transforms = transforms.Compose(
 )
 
 print('Load pre-trained MODNet...')
-pretrained_ckpt = './pretrained/modnet_webcam_portrait_matting.ckpt'
+pretrained_ckpt = args.ckpt_path
 modnet = MODNet(backbone_pretrained=False)
-modnet = nn.DataParallel(modnet).cuda()
-modnet.load_state_dict(torch.load(pretrained_ckpt))
+modnet = nn.DataParallel(modnet)
+modnet.load_state_dict(torch.load(pretrained_ckpt, map_location='cpu'))
+if not args.cpu:
+    modnet.cuda()
 modnet.eval()
 
 print('Init WebCam...')
@@ -38,7 +47,7 @@ while(True):
 
     frame_PIL = Image.fromarray(frame_np)
     frame_tensor = torch_transforms(frame_PIL)
-    frame_tensor = frame_tensor[None, :, :, :].cuda()
+    frame_tensor = frame_tensor[None, :, :, :] if args.cpu else frame_tensor[None, :, :, :].cuda()
 
     with torch.no_grad():
         _, _, matte_tensor = modnet(frame_tensor, inference=True)