lam107 commited on
Commit
71aec6d
·
verified ·
1 Parent(s): 7c385f4

Delete app.py

Browse files
Files changed (1) hide show
  1. app.py +0 -211
app.py DELETED
@@ -1,211 +0,0 @@
1
- #!/usr/bin/env python
2
- # coding: utf-8
3
-
4
- # In[1]:
5
-
6
-
7
-
8
-
9
- # In[2]:
10
-
11
-
12
- get_ipython().system('git clone https://github.com/NVlabs/stylegan2-ada-pytorch.git')
13
- get_ipython().system('cd stylegan2-ada-pytorch')
14
-
15
-
16
- # In[3]:
17
-
18
-
19
- import sys
20
- sys.path.append('/content/stylegan2-ada-pytorch')
21
-
22
- import torch
23
- import pickle
24
- import dnnlib
25
-
26
- # Set device
27
- device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
28
-
29
- # Load the model
30
- pkl_path = "/content/drive/MyDrive/GP_Files_Dona/Experiementation/stylegan2-ffhq-512x512.pkl"
31
-
32
- with open(pkl_path, "rb") as f:
33
- data = pickle.load(f)
34
-
35
- G = data['G_ema'].to(device) # Generator
36
- D = data['D'].to(device) # Discriminator
37
-
38
- # Set discriminator to evaluation mode
39
- D.eval()
40
-
41
- print("StyleGAN2 Discriminator loaded successfully on", device)
42
-
43
-
44
- # In[4]:
45
-
46
-
47
- import torch
48
- import torch.nn as nn
49
- import numpy as np
50
- import torchvision.transforms as transforms
51
- from timm import create_model
52
- import cv2
53
-
54
-
55
- # In[5]:
56
-
57
-
58
- class_names = ["real", "fake"]
59
- model = create_model('swin_large_patch4_window12_384', pretrained=False, num_classes=len(class_names))
60
- model.load_state_dict(torch.load("/content/drive/MyDrive/GP_Files_Yara/swin_large_patch4_window12_384_RGB/best_model_final (1).pth", map_location=device))
61
- model = model.to(device)
62
- model.eval()
63
-
64
-
65
- # In[6]:
66
-
67
-
68
- def compute_gradients(model, pil_image):
69
- transform = transforms.Compose([
70
- transforms.Resize((512, 512)),
71
- transforms.ToTensor(),
72
- transforms.Normalize([0.5], [0.5])
73
- ])
74
- image = transform(pil_image.convert("RGB")).unsqueeze(0).to(device).requires_grad_(True)
75
-
76
- c = torch.zeros([image.shape[0], model.c_dim], device=device) if model.c_dim > 0 else None
77
- output = model(image, c)
78
-
79
- loss = -output.mean()
80
- loss.backward()
81
-
82
- gradients = image.grad.detach().cpu().numpy()
83
- gradients = np.abs(gradients).mean(axis=1)[0]
84
- gradients = (gradients - gradients.min()) / (gradients.max() - gradients.min() + 1e-8)
85
- gradients = (gradients * 255).astype(np.uint8)
86
-
87
- # JET colormap
88
- grad_color = cv2.applyColorMap(gradients, cv2.COLORMAP_JET)
89
- grad_rgb = cv2.cvtColor(grad_color, cv2.COLOR_BGR2RGB)
90
-
91
- return Image.fromarray(grad_rgb)
92
-
93
-
94
- # In[7]:
95
-
96
-
97
- def swin_classifier(pil_image):
98
- transform = transforms.Compose([
99
- transforms.Resize((384, 384)),
100
- transforms.ToTensor(),
101
- transforms.Normalize([0.485, 0.456, 0.406],
102
- [0.229, 0.224, 0.225])
103
- ])
104
- image = transform(pil_image).unsqueeze(0).to(device)
105
-
106
- with torch.no_grad():
107
- output = model(image)
108
- probabilities = torch.softmax(output, dim=1).squeeze().cpu().numpy()
109
- return { "Real": float(probabilities[0]), "Fake": float(probabilities[1]) }
110
-
111
-
112
- # In[8]:
113
-
114
-
115
- def process_image(pil_img):
116
- grad_pil = compute_gradients(D, pil_img).resize((384, 384))
117
- prediction = swin_classifier(grad_pil)
118
- return grad_pil, prediction
119
-
120
-
121
- # In[9]:
122
-
123
-
124
- get_ipython().system('pip install gradio')
125
-
126
-
127
- # In[10]:
128
-
129
-
130
- import gradio as gr
131
- from PIL import Image
132
-
133
-
134
- # In[ ]:
135
-
136
-
137
- with gr.Blocks(theme=gr.themes.Default(primary_hue="purple", secondary_hue="pink")) as demo:
138
- gr.Markdown("**💜 Mukhbir**", elem_id= "logo")
139
- gr.Markdown("# AI Generated Images Detection", elem_id="title")
140
- gr.Markdown("Upload an image to detect whether it is real or fake.", elem_id="desc")
141
-
142
- with gr.Row():
143
- with gr.Column():
144
- image_input = gr.Image(type="pil", label="Upload Image", elem_id="upload-img")
145
- with gr.Row():
146
- clear_button = gr.Button("Clear")
147
- submit_button = gr.Button("Submit", variant="primary")
148
-
149
- with gr.Column():
150
- output_image = gr.Image(label="Gradient Image", type="pil", elem_id="grad-img")
151
- output_label = gr.Label(label="Prediction")
152
-
153
- submit_button.click(fn=process_image, inputs=image_input, outputs=[output_image, output_label])
154
- clear_button.click(fn=lambda: (None, None, None), inputs=None, outputs=[image_input, output_image, output_label])
155
-
156
- # Custom CSS
157
- demo.css = """
158
- #logo, #logo * {
159
- font-size: 20px !important;
160
- top: 10px;
161
- left: 20px;
162
- font-size: 90px;
163
- font-weight: bold;
164
-
165
- }
166
-
167
- #title {
168
- text-align: center;
169
- margin-top: 10px;
170
- }
171
-
172
- #desc {
173
- font-size: 100px;
174
- text-align: left;
175
- }
176
-
177
- #submit-btn button {
178
- background-color: #800080 !important;
179
- color: white !important;
180
- font-size: 18px !important;
181
- padding: 10px 25px !important;
182
- border-radius: 8px !important;
183
- }
184
-
185
- #submit-btn button:hover {
186
- background-color: #6A006A !important;
187
- }
188
-
189
- #clear-btn button {
190
- font-size: 18px !important;
191
- padding: 10px 25px !important;
192
- border-radius: 8px !important;
193
- }
194
-
195
- #upload-img img {
196
- max-width: 256px;
197
- max-height: 256px;
198
- object-fit: contain;
199
- }
200
-
201
-
202
- #grad-img img {
203
- max-width: 200px;
204
- max-height: 200px;
205
- border-radius: 10px;
206
- }
207
-
208
- """
209
-
210
- demo.launch(share=True, debug=True)
211
-