Spaces:
Runtime error
Runtime error
danielsapit
commited on
Commit
•
3ae9a61
1
Parent(s):
5cb71e1
Update app.py
Browse files
app.py
CHANGED
@@ -20,6 +20,8 @@ for model_path in ['fbcnn_gray.pth','fbcnn_color.pth']:
|
|
20 |
|
21 |
def inference(input_img, is_gray, input_quality, zoom, x_shift, y_shift):
|
22 |
|
|
|
|
|
23 |
if is_gray:
|
24 |
n_channels = 1 # set 1 for grayscale image, set 3 for color image
|
25 |
model_name = 'fbcnn_gray.pth'
|
@@ -35,7 +37,7 @@ def inference(input_img, is_gray, input_quality, zoom, x_shift, y_shift):
|
|
35 |
model_path = model_name
|
36 |
|
37 |
if os.path.exists(model_path):
|
38 |
-
print(f'
|
39 |
else:
|
40 |
print("downloading model")
|
41 |
os.makedirs(os.path.dirname(model_path), exist_ok=True)
|
@@ -44,10 +46,14 @@ def inference(input_img, is_gray, input_quality, zoom, x_shift, y_shift):
|
|
44 |
open(model_path, 'wb').write(r.content)
|
45 |
|
46 |
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
|
|
47 |
|
48 |
# ----------------------------------------
|
49 |
# load model
|
50 |
# ----------------------------------------
|
|
|
|
|
|
|
51 |
model = net(in_nc=n_channels, out_nc=n_channels, nc=nc, nb=nb, act_mode='R')
|
52 |
model.load_state_dict(torch.load(model_path), strict=True)
|
53 |
model.eval()
|
@@ -55,6 +61,8 @@ def inference(input_img, is_gray, input_quality, zoom, x_shift, y_shift):
|
|
55 |
v.requires_grad = False
|
56 |
model = model.to(device)
|
57 |
|
|
|
|
|
58 |
test_results = OrderedDict()
|
59 |
test_results['psnr'] = []
|
60 |
test_results['ssim'] = []
|
@@ -64,6 +72,7 @@ def inference(input_img, is_gray, input_quality, zoom, x_shift, y_shift):
|
|
64 |
# (1) img_L
|
65 |
# ------------------------------------
|
66 |
|
|
|
67 |
if n_channels == 1:
|
68 |
open_cv_image = Image.fromarray(input_img)
|
69 |
open_cv_image = ImageOps.grayscale(open_cv_image)
|
@@ -76,21 +85,31 @@ def inference(input_img, is_gray, input_quality, zoom, x_shift, y_shift):
|
|
76 |
else:
|
77 |
open_cv_image = cv2.cvtColor(open_cv_image, cv2.COLOR_BGR2RGB) # RGB
|
78 |
|
|
|
79 |
img_L = util.uint2tensor4(open_cv_image)
|
|
|
|
|
80 |
img_L = img_L.to(device)
|
81 |
|
82 |
# ------------------------------------
|
83 |
# (2) img_E
|
84 |
# ------------------------------------
|
85 |
-
|
|
|
86 |
img_E,QF = model(img_L)
|
|
|
87 |
img_E = util.tensor2single(img_E)
|
|
|
88 |
img_E = util.single2uint(img_E)
|
89 |
-
|
|
|
90 |
qf_input = torch.tensor([[1-input_quality/100]]).cuda() if device == torch.device('cuda') else torch.tensor([[1-input_quality/100]])
|
|
|
91 |
img_E,QF = model(img_L, qf_input)
|
92 |
|
|
|
93 |
img_E = util.tensor2single(img_E)
|
|
|
94 |
img_E = util.single2uint(img_E)
|
95 |
|
96 |
if img_E.ndim == 3:
|
@@ -111,7 +130,9 @@ def inference(input_img, is_gray, input_quality, zoom, x_shift, y_shift):
|
|
111 |
in_img = in_img.resize((int(zoom_w/zoom), int(zoom_h/zoom)), Image.NEAREST)
|
112 |
out_img = out_img.crop((zoom_left, zoom_top, zoom_right, zoom_bottom))
|
113 |
out_img = out_img.resize((int(zoom_w/zoom), int(zoom_h/zoom)), Image.NEAREST)
|
114 |
-
|
|
|
|
|
115 |
return img_E, in_img, out_img
|
116 |
|
117 |
gr.Interface(
|
|
|
20 |
|
21 |
def inference(input_img, is_gray, input_quality, zoom, x_shift, y_shift):
|
22 |
|
23 |
+
print("img size:",Image.fromarray(input_img).size)
|
24 |
+
|
25 |
if is_gray:
|
26 |
n_channels = 1 # set 1 for grayscale image, set 3 for color image
|
27 |
model_name = 'fbcnn_gray.pth'
|
|
|
37 |
model_path = model_name
|
38 |
|
39 |
if os.path.exists(model_path):
|
40 |
+
print(f'{model_path} already exists.')
|
41 |
else:
|
42 |
print("downloading model")
|
43 |
os.makedirs(os.path.dirname(model_path), exist_ok=True)
|
|
|
46 |
open(model_path, 'wb').write(r.content)
|
47 |
|
48 |
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
49 |
+
print("device:",device)
|
50 |
|
51 |
# ----------------------------------------
|
52 |
# load model
|
53 |
# ----------------------------------------
|
54 |
+
|
55 |
+
print(f'loading model from {model_path}')
|
56 |
+
|
57 |
model = net(in_nc=n_channels, out_nc=n_channels, nc=nc, nb=nb, act_mode='R')
|
58 |
model.load_state_dict(torch.load(model_path), strict=True)
|
59 |
model.eval()
|
|
|
61 |
v.requires_grad = False
|
62 |
model = model.to(device)
|
63 |
|
64 |
+
print("Model loaded.")
|
65 |
+
|
66 |
test_results = OrderedDict()
|
67 |
test_results['psnr'] = []
|
68 |
test_results['ssim'] = []
|
|
|
72 |
# (1) img_L
|
73 |
# ------------------------------------
|
74 |
|
75 |
+
print("#if n_channels")
|
76 |
if n_channels == 1:
|
77 |
open_cv_image = Image.fromarray(input_img)
|
78 |
open_cv_image = ImageOps.grayscale(open_cv_image)
|
|
|
85 |
else:
|
86 |
open_cv_image = cv2.cvtColor(open_cv_image, cv2.COLOR_BGR2RGB) # RGB
|
87 |
|
88 |
+
print("#util.uint2tensor4(open_cv_image)")
|
89 |
img_L = util.uint2tensor4(open_cv_image)
|
90 |
+
|
91 |
+
print("#img_L.to(device)")
|
92 |
img_L = img_L.to(device)
|
93 |
|
94 |
# ------------------------------------
|
95 |
# (2) img_E
|
96 |
# ------------------------------------
|
97 |
+
|
98 |
+
print("#model(img_L)")
|
99 |
img_E,QF = model(img_L)
|
100 |
+
print("#util.tensor2single(img_E)")
|
101 |
img_E = util.tensor2single(img_E)
|
102 |
+
print("#util.single2uint(img_E)")
|
103 |
img_E = util.single2uint(img_E)
|
104 |
+
|
105 |
+
print("#torch.tensor([[1-input_quality/100]]).cuda() || torch.tensor([[1-input_quality/100]])")
|
106 |
qf_input = torch.tensor([[1-input_quality/100]]).cuda() if device == torch.device('cuda') else torch.tensor([[1-input_quality/100]])
|
107 |
+
print("#util.single2uint(img_E)")
|
108 |
img_E,QF = model(img_L, qf_input)
|
109 |
|
110 |
+
print("#util.tensor2single(img_E)")
|
111 |
img_E = util.tensor2single(img_E)
|
112 |
+
print("#util.single2uint(img_E)")
|
113 |
img_E = util.single2uint(img_E)
|
114 |
|
115 |
if img_E.ndim == 3:
|
|
|
130 |
in_img = in_img.resize((int(zoom_w/zoom), int(zoom_h/zoom)), Image.NEAREST)
|
131 |
out_img = out_img.crop((zoom_left, zoom_top, zoom_right, zoom_bottom))
|
132 |
out_img = out_img.resize((int(zoom_w/zoom), int(zoom_h/zoom)), Image.NEAREST)
|
133 |
+
|
134 |
+
print("--generating preview finished")
|
135 |
+
|
136 |
return img_E, in_img, out_img
|
137 |
|
138 |
gr.Interface(
|