Update app.py
Browse files
app.py
CHANGED
@@ -1,151 +1,3 @@
|
|
1 |
-
|
2 |
-
|
3 |
-
|
4 |
-
from facenet_pytorch import MTCNN
|
5 |
-
from torchvision import transforms
|
6 |
-
import torch, PIL
|
7 |
-
from tqdm.notebook import tqdm
|
8 |
-
import gradio as gr
|
9 |
-
import torch
|
10 |
-
|
11 |
-
modelarcanev4 = hf_hub_download(repo_id="akhaliq/ArcaneGANv0.4", filename="ArcaneGANv0.4.jit")
|
12 |
-
modelarcanev3 = hf_hub_download(repo_id="akhaliq/ArcaneGANv0.3", filename="ArcaneGANv0.3.jit")
|
13 |
-
modelarcanev2 = hf_hub_download(repo_id="akhaliq/ArcaneGANv0.2", filename="ArcaneGANv0.2.jit")
|
14 |
-
|
15 |
-
|
16 |
-
mtcnn = MTCNN(image_size=256, margin=80)
|
17 |
-
|
18 |
-
# simplest ye olde trustworthy MTCNN for face detection with landmarks
|
19 |
-
def detect(img):
|
20 |
-
|
21 |
-
# Detect faces
|
22 |
-
batch_boxes, batch_probs, batch_points = mtcnn.detect(img, landmarks=True)
|
23 |
-
# Select faces
|
24 |
-
if not mtcnn.keep_all:
|
25 |
-
batch_boxes, batch_probs, batch_points = mtcnn.select_boxes(
|
26 |
-
batch_boxes, batch_probs, batch_points, img, method=mtcnn.selection_method
|
27 |
-
)
|
28 |
-
|
29 |
-
return batch_boxes, batch_points
|
30 |
-
|
31 |
-
# my version of isOdd, should make a separate repo for it :D
|
32 |
-
def makeEven(_x):
|
33 |
-
return _x if (_x % 2 == 0) else _x+1
|
34 |
-
|
35 |
-
# the actual scaler function
|
36 |
-
def scale(boxes, _img, max_res=1_500_000, target_face=256, fixed_ratio=0, max_upscale=2, VERBOSE=False):
|
37 |
-
|
38 |
-
x, y = _img.size
|
39 |
-
|
40 |
-
ratio = 2 #initial ratio
|
41 |
-
|
42 |
-
#scale to desired face size
|
43 |
-
if (boxes is not None):
|
44 |
-
if len(boxes)>0:
|
45 |
-
ratio = target_face/max(boxes[0][2:]-boxes[0][:2]);
|
46 |
-
ratio = min(ratio, max_upscale)
|
47 |
-
if VERBOSE: print('up by', ratio)
|
48 |
-
|
49 |
-
if fixed_ratio>0:
|
50 |
-
if VERBOSE: print('fixed ratio')
|
51 |
-
ratio = fixed_ratio
|
52 |
-
|
53 |
-
x*=ratio
|
54 |
-
y*=ratio
|
55 |
-
|
56 |
-
#downscale to fit into max res
|
57 |
-
res = x*y
|
58 |
-
if res > max_res:
|
59 |
-
ratio = pow(res/max_res,1/2);
|
60 |
-
if VERBOSE: print(ratio)
|
61 |
-
x=int(x/ratio)
|
62 |
-
y=int(y/ratio)
|
63 |
-
|
64 |
-
#make dimensions even, because usually NNs fail on uneven dimensions due skip connection size mismatch
|
65 |
-
x = makeEven(int(x))
|
66 |
-
y = makeEven(int(y))
|
67 |
-
|
68 |
-
size = (x, y)
|
69 |
-
|
70 |
-
return _img.resize(size)
|
71 |
-
|
72 |
-
"""
|
73 |
-
A useful scaler algorithm, based on face detection.
|
74 |
-
Takes PIL.Image, returns a uniformly scaled PIL.Image
|
75 |
-
boxes: a list of detected bboxes
|
76 |
-
_img: PIL.Image
|
77 |
-
max_res: maximum pixel area to fit into. Use to stay below the VRAM limits of your GPU.
|
78 |
-
target_face: desired face size. Upscale or downscale the whole image to fit the detected face into that dimension.
|
79 |
-
fixed_ratio: fixed scale. Ignores the face size, but doesn't ignore the max_res limit.
|
80 |
-
max_upscale: maximum upscale ratio. Prevents from scaling images with tiny faces to a blurry mess.
|
81 |
-
"""
|
82 |
-
|
83 |
-
def scale_by_face_size(_img, max_res=1_500_000, target_face=256, fix_ratio=0, max_upscale=2, VERBOSE=False):
|
84 |
-
boxes = None
|
85 |
-
boxes, _ = detect(_img)
|
86 |
-
if VERBOSE: print('boxes',boxes)
|
87 |
-
img_resized = scale(boxes, _img, max_res, target_face, fix_ratio, max_upscale, VERBOSE)
|
88 |
-
return img_resized
|
89 |
-
|
90 |
-
|
91 |
-
size = 256
|
92 |
-
|
93 |
-
means = [0.485, 0.456, 0.406]
|
94 |
-
stds = [0.229, 0.224, 0.225]
|
95 |
-
|
96 |
-
t_stds = torch.tensor(stds).cuda().half()[:,None,None]
|
97 |
-
t_means = torch.tensor(means).cuda().half()[:,None,None]
|
98 |
-
|
99 |
-
def makeEven(_x):
|
100 |
-
return int(_x) if (_x % 2 == 0) else int(_x+1)
|
101 |
-
|
102 |
-
img_transforms = transforms.Compose([
|
103 |
-
transforms.ToTensor(),
|
104 |
-
transforms.Normalize(means,stds)])
|
105 |
-
|
106 |
-
def tensor2im(var):
|
107 |
-
return var.mul(t_stds).add(t_means).mul(255.).clamp(0,255).permute(1,2,0)
|
108 |
-
|
109 |
-
def proc_pil_img(input_image, model):
|
110 |
-
transformed_image = img_transforms(input_image)[None,...].cuda().half()
|
111 |
-
|
112 |
-
with torch.no_grad():
|
113 |
-
result_image = model(transformed_image)[0]
|
114 |
-
output_image = tensor2im(result_image)
|
115 |
-
output_image = output_image.detach().cpu().numpy().astype('uint8')
|
116 |
-
output_image = PIL.Image.fromarray(output_image)
|
117 |
-
return output_image
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
modelv4 = torch.jit.load(modelarcanev4).eval().cuda().half()
|
122 |
-
modelv3 = torch.jit.load(modelarcanev3).eval().cuda().half()
|
123 |
-
modelv2 = torch.jit.load(modelarcanev2).eval().cuda().half()
|
124 |
-
|
125 |
-
def process(im, version):
|
126 |
-
if version == 'version 0.4':
|
127 |
-
im = scale_by_face_size(im, target_face=256, max_res=1_500_000, max_upscale=1)
|
128 |
-
res = proc_pil_img(im, modelv4)
|
129 |
-
elif version == 'version 0.3':
|
130 |
-
im = scale_by_face_size(im, target_face=256, max_res=1_500_000, max_upscale=1)
|
131 |
-
res = proc_pil_img(im, modelv3)
|
132 |
-
else:
|
133 |
-
im = scale_by_face_size(im, target_face=256, max_res=1_500_000, max_upscale=1)
|
134 |
-
res = proc_pil_img(im, modelv2)
|
135 |
-
return res
|
136 |
-
|
137 |
-
title = "ArcaneGAN"
|
138 |
-
description = "Gradio demo for ArcaneGAN, portrait to Arcane style. To use it, simply upload your image, or click one of the examples to load them. Read more at the links below."
|
139 |
-
article = "<div style='text-align: center;'>ArcaneGan by <a href='https://twitter.com/devdef' target='_blank'>Alexander S</a> | <a href='https://github.com/Sxela/ArcaneGAN' target='_blank'>Github Repo</a> | <center><img src='https://visitor-badge.glitch.me/badge?page_id=akhaliq_arcanegan' alt='visitor badge'></center></div>"
|
140 |
-
|
141 |
-
gr.Interface(
|
142 |
-
process,
|
143 |
-
[gr.inputs.Image(source="webcam", type="pil", label="Input"),gr.inputs.Radio(choices=['version 0.2','version 0.3','version 0.4'], type="value", default='version 0.4', label='version')
|
144 |
-
],
|
145 |
-
gr.outputs.Image(type="pil", label="Output"),
|
146 |
-
title=title,
|
147 |
-
description=description,
|
148 |
-
article=article,
|
149 |
-
allow_flagging=False,
|
150 |
-
allow_screenshot=False
|
151 |
-
).launch()
|
|
|
1 |
+
gr.Interface.load(
|
2 |
+
"spaces/abidlabs/remove-bg", inputs="webcam", title="Remove your webcam background!"
|
3 |
+
).launch()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|