Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -5,34 +5,14 @@ import numpy as np
|
|
5 |
import cv2
|
6 |
import gradio as gr
|
7 |
import os
|
8 |
-
import subprocess
|
9 |
-
import gdown
|
10 |
-
|
11 |
-
os.makedirs('/content', exist_ok=True)
|
12 |
-
# Download model
|
13 |
-
# Pix2pix model
|
14 |
-
model_url = 'https://drive.google.com/drive/folders/1jOxiKyf8n7fwNZfgeZyUrNJ90LEmIL3S?usp=sharing'
|
15 |
-
os.makedirs('/content/pix2pix', exist_ok=True)
|
16 |
-
subprocess.run(['gdown', '--fuzzy', model_url, '-O', '/content/pix2pix', '--folder'], check=True)
|
17 |
-
|
18 |
-
# WD-Net
|
19 |
-
model_url = 'https://drive.google.com/file/d/1M8EOE4Ej8oS4_0BHCEwExxu5CMFS5HQZ/view?usp=sharing'
|
20 |
-
os.makedirs('/content/WD-Net', exist_ok=True)
|
21 |
-
subprocess.run(['gdown', '--fuzzy', model_url, '-O', '/content/WD-Net/model.zip'], check=True)
|
22 |
-
subprocess.run(['unzip', '/content/WD-Net/model.zip', '-d', '/content/WD-Net'], check=True)
|
23 |
-
|
24 |
-
# MS-UNet
|
25 |
-
model_url = 'https://drive.google.com/file/d/1-0_bEWTItkILbCJQ4ViEBGg0zJPaIcC1/view?usp=sharing'
|
26 |
-
os.makedirs('/content/MS-UNet', exist_ok=True)
|
27 |
-
subprocess.run(['gdown', '--fuzzy', model_url, '-O', '/content/MS-UNet/unet.keras'], check=True)
|
28 |
|
29 |
# Load model
|
30 |
# Load Pix2Pix
|
31 |
-
pix2pix_path = '
|
32 |
Pix2Pix = keras.saving.load_model(pix2pix_path)
|
33 |
|
34 |
# Load MS-UNet
|
35 |
-
unet_path = '
|
36 |
MS_UNet = keras.saving.load_model(unet_path)
|
37 |
|
38 |
# Load WD-Net
|
@@ -43,16 +23,66 @@ class Clip(keras.layers.Layer):
|
|
43 |
def call(self, input):
|
44 |
return tf.clip_by_value(input, 0, 1)
|
45 |
|
46 |
-
|
47 |
-
|
|
|
|
|
48 |
|
49 |
# Define infer function
|
50 |
def infer(img, model='WD-Net'):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
51 |
# Image original shape
|
52 |
org_shape = img.shape
|
|
|
|
|
|
|
|
|
53 |
# Choose model
|
54 |
if model == 'WD-Net':
|
55 |
-
generator =
|
56 |
img = tf.image.resize(img, [256, 256], method='area')
|
57 |
# Normalize image and return
|
58 |
img = tf.cast(img, tf.float32) / 255.
|
@@ -81,15 +111,28 @@ def infer(img, model='WD-Net'):
|
|
81 |
rm_wt = rm_wt[0]
|
82 |
rm_wt = cv2.resize(rm_wt, (org_shape[1], org_shape[0]))
|
83 |
out_img = (rm_wt * 255).astype(np.uint8)
|
84 |
-
return out_img
|
85 |
|
86 |
# Main gradio code
|
|
|
|
|
|
|
|
|
|
|
87 |
model_list = ['WD-Net', 'MS-UNet', 'Pix2Pix']
|
88 |
|
89 |
demo = gr.Interface(
|
90 |
fn=infer,
|
91 |
-
inputs=[gr.Image(), gr.Dropdown(model_list)],
|
92 |
-
outputs=gr.Image(),
|
93 |
)
|
94 |
|
95 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
import cv2
|
6 |
import gradio as gr
|
7 |
import os
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
|
9 |
# Load model
|
10 |
# Load Pix2Pix
|
11 |
+
pix2pix_path = './model/wt_generator_best.keras'
|
12 |
Pix2Pix = keras.saving.load_model(pix2pix_path)
|
13 |
|
14 |
# Load MS-UNet
|
15 |
+
unet_path = './model/unet.keras'
|
16 |
MS_UNet = keras.saving.load_model(unet_path)
|
17 |
|
18 |
# Load WD-Net
|
|
|
23 |
def call(self, input):
|
24 |
return tf.clip_by_value(input, 0, 1)
|
25 |
|
26 |
+
old_gen_path = './model/generator_epoch_10.keras'
|
27 |
+
WD_Net_old = keras.saving.load_model(old_gen_path)
|
28 |
+
new_gen_path = './model/WD-Net_generator.keras'
|
29 |
+
WD_Net_new = keras.saving.load_model(new_gen_path)
|
30 |
|
31 |
# Define infer function
|
32 |
def infer(img, model='WD-Net'):
|
33 |
+
# Read image
|
34 |
+
# img = tf.image.decode_png(tf.io.read_file('./data/' + img_path), channels=3)
|
35 |
+
# Image original shape
|
36 |
+
org_shape = img.shape
|
37 |
+
org_img = tf.image.resize(img, [256, 256], method='area')
|
38 |
+
org_img = tf.cast(org_img, tf.uint8).numpy()
|
39 |
+
org_img = cv2.resize(org_img, (org_shape[1], org_shape[0]))
|
40 |
+
|
41 |
+
# Choose model
|
42 |
+
if model == 'WD-Net':
|
43 |
+
generator = WD_Net_old
|
44 |
+
img = tf.image.resize(img, [256, 256], method='area')
|
45 |
+
# Normalize image and return
|
46 |
+
img = tf.cast(img, tf.float32) / 255.
|
47 |
+
img = tf.expand_dims(img, axis=0)
|
48 |
+
rm_wt = generator.predict(img, verbose=0)
|
49 |
+
rm_wt = rm_wt['I'][0]
|
50 |
+
rm_wt = cv2.resize(rm_wt, (org_shape[1], org_shape[0]))
|
51 |
+
out_img = (rm_wt * 255).astype(np.uint8)
|
52 |
+
elif model == 'MS-UNet':
|
53 |
+
generator = MS_UNet
|
54 |
+
img = tf.image.resize(img, [256, 256], method='area')
|
55 |
+
# Normalize image and return
|
56 |
+
img = (tf.cast(img, tf.float32) - 127.5) / 127.5
|
57 |
+
img = tf.expand_dims(img, axis=0)
|
58 |
+
rm_wt = generator.predict(img, verbose=0)
|
59 |
+
rm_wt = rm_wt[0]
|
60 |
+
rm_wt = cv2.resize(rm_wt, (org_shape[1], org_shape[0]))
|
61 |
+
out_img = ((rm_wt + 1) / 2 * 255).astype(np.uint8)
|
62 |
+
elif model == 'Pix2Pix':
|
63 |
+
generator = Pix2Pix
|
64 |
+
img = tf.image.resize(img, [256, 256], method='area')
|
65 |
+
# Normalize image and return
|
66 |
+
img = tf.cast(img, tf.float32) / 255.
|
67 |
+
img = tf.expand_dims(img, axis=0)
|
68 |
+
rm_wt = generator.predict(img, verbose=0)
|
69 |
+
rm_wt = rm_wt[0]
|
70 |
+
rm_wt = cv2.resize(rm_wt, (org_shape[1], org_shape[0]))
|
71 |
+
out_img = (rm_wt * 255).astype(np.uint8)
|
72 |
+
return org_img, out_img
|
73 |
+
|
74 |
+
def infer_v1(img_path, model="WD_Net"):
|
75 |
+
# Read image
|
76 |
+
img = tf.image.decode_png(tf.io.read_file('./data/' + img_path), channels=3)
|
77 |
# Image original shape
|
78 |
org_shape = img.shape
|
79 |
+
# org_img = tf.image.resize(img, [256, 256], method='area')
|
80 |
+
org_img = tf.cast(img, tf.uint8).numpy()
|
81 |
+
# org_img = cv2.resize(org_img, (org_shape[1], org_shape[0]))
|
82 |
+
|
83 |
# Choose model
|
84 |
if model == 'WD-Net':
|
85 |
+
generator = WD_Net_new
|
86 |
img = tf.image.resize(img, [256, 256], method='area')
|
87 |
# Normalize image and return
|
88 |
img = tf.cast(img, tf.float32) / 255.
|
|
|
111 |
rm_wt = rm_wt[0]
|
112 |
rm_wt = cv2.resize(rm_wt, (org_shape[1], org_shape[0]))
|
113 |
out_img = (rm_wt * 255).astype(np.uint8)
|
114 |
+
return org_img, out_img
|
115 |
|
116 |
# Main gradio code
|
117 |
+
# Define data and sort it
|
118 |
+
data = os.listdir('./data')
|
119 |
+
data.sort()
|
120 |
+
|
121 |
+
# Model list
|
122 |
model_list = ['WD-Net', 'MS-UNet', 'Pix2Pix']
|
123 |
|
124 |
demo = gr.Interface(
|
125 |
fn=infer,
|
126 |
+
inputs=[gr.Image(label="Choose an Image"), gr.Dropdown(model_list, label="Model")],
|
127 |
+
outputs=[gr.Image(label="Watermarked Image"), gr.Image(label="Removed Watermarked Image")],
|
128 |
)
|
129 |
|
130 |
+
demo_v1 = gr.Interface(
|
131 |
+
fn=infer_v1,
|
132 |
+
inputs=[gr.Dropdown(data, label="Choose an Image"), gr.Dropdown(model_list, label="Model")],
|
133 |
+
outputs=[gr.Image(label="Watermarked Image"), gr.Image(label="Removed Watermarked Image")],
|
134 |
+
)
|
135 |
+
|
136 |
+
tabbed_interface = gr.TabbedInterface([demo, demo_v1], ["Document", "Patch"], title="Watermark Removal")
|
137 |
+
|
138 |
+
tabbed_interface.launch()
|