Spaces:
Running
Running
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,75 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import cv2
|
3 |
+
import torch
|
4 |
+
import numpy as np
|
5 |
+
from torchvision import transforms
|
6 |
+
|
7 |
+
description = "Automatically remove the image background from a profile photo. Based on a [Space by eugenesiow](https://huggingface.co/spaces/eugenesiow/remove-bg)."
|
8 |
+
|
9 |
+
|
10 |
+
def make_transparent_foreground(pic, mask):
|
11 |
+
# split the image into channels
|
12 |
+
b, g, r = cv2.split(np.array(pic).astype('uint8'))
|
13 |
+
# add an alpha channel with and fill all with transparent pixels (max 255)
|
14 |
+
a = np.ones(mask.shape, dtype='uint8') * 255
|
15 |
+
# merge the alpha channel back
|
16 |
+
alpha_im = cv2.merge([b, g, r, a], 4)
|
17 |
+
# create a transparent background
|
18 |
+
bg = np.zeros(alpha_im.shape)
|
19 |
+
# setup the new mask
|
20 |
+
new_mask = np.stack([mask, mask, mask, mask], axis=2)
|
21 |
+
# copy only the foreground color pixels from the original image where mask is set
|
22 |
+
foreground = np.where(new_mask, alpha_im, bg).astype(np.uint8)
|
23 |
+
|
24 |
+
return foreground
|
25 |
+
|
26 |
+
|
27 |
+
def remove_background(input_image):
|
28 |
+
preprocess = transforms.Compose([
|
29 |
+
transforms.ToTensor(),
|
30 |
+
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
|
31 |
+
])
|
32 |
+
|
33 |
+
input_tensor = preprocess(input_image)
|
34 |
+
input_batch = input_tensor.unsqueeze(0) # create a mini-batch as expected by the model
|
35 |
+
|
36 |
+
# move the input and model to GPU for speed if available
|
37 |
+
if torch.cuda.is_available():
|
38 |
+
input_batch = input_batch.to('cuda')
|
39 |
+
model.to('cuda')
|
40 |
+
|
41 |
+
with torch.no_grad():
|
42 |
+
output = model(input_batch)['out'][0]
|
43 |
+
output_predictions = output.argmax(0)
|
44 |
+
|
45 |
+
# create a binary (black and white) mask of the profile foreground
|
46 |
+
mask = output_predictions.byte().cpu().numpy()
|
47 |
+
background = np.zeros(mask.shape)
|
48 |
+
bin_mask = np.where(mask, 255, background).astype(np.uint8)
|
49 |
+
|
50 |
+
foreground = make_transparent_foreground(input_image, bin_mask)
|
51 |
+
|
52 |
+
return foreground, bin_mask
|
53 |
+
|
54 |
+
|
55 |
+
def inference(img):
|
56 |
+
foreground, _ = remove_background(img)
|
57 |
+
return foreground
|
58 |
+
|
59 |
+
|
60 |
+
torch.hub.download_url_to_file('https://pbs.twimg.com/profile_images/691700243809718272/z7XZUARB_400x400.jpg',
|
61 |
+
'demis.jpg')
|
62 |
+
torch.hub.download_url_to_file('https://hai.stanford.edu/sites/default/files/styles/person_medium/public/2020-03/hai_1512feifei.png?itok=INFuLABp',
|
63 |
+
'lifeifei.png')
|
64 |
+
model = torch.hub.load('pytorch/vision:v0.6.0', 'deeplabv3_resnet101', pretrained=True)
|
65 |
+
model.eval()
|
66 |
+
|
67 |
+
gr.Interface(
|
68 |
+
inference,
|
69 |
+
gr.Image(type="pil", label="Input"),
|
70 |
+
gr.Image(type="pil", label="Output"),
|
71 |
+
description=description,
|
72 |
+
examples=[['demis.jpg'], ['lifeifei.png']],
|
73 |
+
enable_queue=True,
|
74 |
+
css=".footer{display:none !important}"
|
75 |
+
).launch(debug=False)
|