Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -10,9 +10,12 @@ MODEL_IDENTIFIER = r"Ateeqq/ai-vs-human-image-detector"
|
|
10 |
DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
11 |
|
12 |
# --- Suppress specific warnings ---
|
|
|
13 |
warnings.filterwarnings("ignore", message="Possibly corrupt EXIF data.")
|
|
|
14 |
warnings.filterwarnings("ignore", message=".*You are using the default legacy behaviour.*")
|
15 |
|
|
|
16 |
# --- Load Model and Processor (Load once at startup) ---
|
17 |
print(f"Using device: {DEVICE}")
|
18 |
print(f"Loading processor from: {MODEL_IDENTIFIER}")
|
@@ -25,184 +28,93 @@ try:
|
|
25 |
print("Model and processor loaded successfully.")
|
26 |
except Exception as e:
|
27 |
print(f"FATAL: Error loading model or processor: {e}")
|
|
|
28 |
raise gr.Error(f"Failed to load the model: {e}. Cannot start the application.") from e
|
29 |
|
30 |
# --- Prediction Function ---
|
31 |
def classify_image(image_pil):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
if image_pil is None:
|
|
|
33 |
print("Warning: No image provided.")
|
34 |
-
return {}
|
35 |
|
36 |
print("Processing image...")
|
37 |
try:
|
|
|
38 |
image = image_pil.convert("RGB")
|
|
|
|
|
39 |
inputs = processor(images=image, return_tensors="pt").to(DEVICE)
|
40 |
|
|
|
41 |
print("Running inference...")
|
42 |
with torch.no_grad():
|
43 |
outputs = model(**inputs)
|
44 |
logits = outputs.logits
|
45 |
|
46 |
-
probabilities
|
|
|
|
|
|
|
|
|
47 |
results = {}
|
48 |
for i, prob in enumerate(probabilities):
|
49 |
label = model.config.id2label[i]
|
50 |
-
results[label] =
|
51 |
|
52 |
print(f"Prediction results: {results}")
|
53 |
return results
|
|
|
54 |
except Exception as e:
|
55 |
print(f"Error during prediction: {e}")
|
56 |
-
|
|
|
|
|
57 |
|
58 |
-
# ---
|
|
|
|
|
|
|
59 |
example_dir = "examples"
|
60 |
example_images = []
|
61 |
-
if os.path.exists(example_dir)
|
62 |
for img_name in os.listdir(example_dir):
|
63 |
if img_name.lower().endswith(('.png', '.jpg', '.jpeg', '.webp')):
|
64 |
-
|
65 |
-
|
66 |
-
print(f"Found examples: {example_images}")
|
67 |
-
else:
|
68 |
-
print("No valid image files found in 'examples' directory.")
|
69 |
else:
|
70 |
-
print("No 'examples' directory found
|
71 |
-
|
72 |
-
|
73 |
-
#
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
font-weight: bold;
|
81 |
-
font-size: 2.5em;
|
82 |
-
margin-bottom: 5px;
|
83 |
-
/* color removed - let theme handle */
|
84 |
-
}
|
85 |
-
|
86 |
-
/* Style the description */
|
87 |
-
#app-description {
|
88 |
-
text-align: center;
|
89 |
-
font-size: 1.1em;
|
90 |
-
margin-bottom: 25px;
|
91 |
-
/* color removed - let theme handle */
|
92 |
-
}
|
93 |
-
#app-description code { /* Style model name - theme might handle this, but can force */
|
94 |
-
font-weight: bold;
|
95 |
-
background-color: rgba(255, 255, 255, 0.1); /* Slightly lighter background for code */
|
96 |
-
padding: 2px 5px;
|
97 |
-
border-radius: 4px;
|
98 |
-
color: #c5f7dc; /* Light green text for code block */
|
99 |
-
}
|
100 |
-
#app-description strong { /* Style device name */
|
101 |
-
color: #2dd4bf; /* Brighter teal/emerald for dark theme */
|
102 |
-
font-weight: bold;
|
103 |
-
}
|
104 |
-
|
105 |
-
/* Style the results heading */
|
106 |
-
#results-heading {
|
107 |
-
text-align: center;
|
108 |
-
font-size: 1.2em;
|
109 |
-
margin-bottom: 10px;
|
110 |
-
/* color removed - let theme handle */
|
111 |
-
}
|
112 |
-
|
113 |
-
/* Add some definition to input/output columns if needed */
|
114 |
-
#input-column, #output-column {
|
115 |
-
border: 1px solid #4b5563; /* Darker border for dark theme */
|
116 |
-
border-radius: 12px;
|
117 |
-
padding: 20px;
|
118 |
-
box-shadow: 0 2px 8px rgba(0, 0, 0, 0.1); /* Subtle shadow, works on dark too */
|
119 |
-
/* background-color removed - let theme handle */
|
120 |
-
}
|
121 |
-
|
122 |
-
/* Ensure label text inside columns is readable */
|
123 |
-
#prediction-label .label-name { font-weight: bold; font-size: 1.1em; }
|
124 |
-
#prediction-label .confidence { font-size: 1em; }
|
125 |
-
|
126 |
-
|
127 |
-
/* Footer styling */
|
128 |
-
#app-footer {
|
129 |
-
margin-top: 40px;
|
130 |
-
padding-top: 20px;
|
131 |
-
border-top: 1px solid #374151; /* Darker border for footer */
|
132 |
-
text-align: center;
|
133 |
-
font-size: 0.9em;
|
134 |
-
/* color removed - let theme handle */
|
135 |
-
}
|
136 |
-
#app-footer a {
|
137 |
-
color: #60a5fa; /* Lighter blue for links */
|
138 |
-
text-decoration: none;
|
139 |
-
}
|
140 |
-
#app-footer a:hover {
|
141 |
-
text-decoration: underline;
|
142 |
-
}
|
143 |
-
"""
|
144 |
-
|
145 |
-
# --- Gradio Interface using Blocks and Theme ---
|
146 |
-
# Use the theme string identifier for the dark mode variant
|
147 |
-
# Other options: "default/dark", "monochrome/dark", "glass/dark"
|
148 |
-
with gr.Blocks(theme="soft/dark", css=css) as iface: # <<< CHANGE IS HERE
|
149 |
-
# Title and Description
|
150 |
-
gr.Markdown("# AI vs Human Image Detector", elem_id="app-title")
|
151 |
-
gr.Markdown(
|
152 |
f"Upload an image to classify if it was likely generated by AI or created by a human. "
|
153 |
-
f"Uses the `{MODEL_IDENTIFIER}` model. Running on **{str(DEVICE).upper()}**."
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
|
163 |
-
|
164 |
-
|
165 |
-
|
166 |
-
submit_button = gr.Button("🔍 Classify Image", variant="primary")
|
167 |
-
|
168 |
-
with gr.Column(scale=1, min_width=300, elem_id="output-column"):
|
169 |
-
gr.Markdown("📊 **Prediction Results**", elem_id="results-heading")
|
170 |
-
result_output = gr.Label(
|
171 |
-
num_top_classes=2,
|
172 |
-
label="Classification",
|
173 |
-
elem_id="prediction-label"
|
174 |
-
)
|
175 |
-
|
176 |
-
# Examples Section
|
177 |
-
if example_images:
|
178 |
-
gr.Examples(
|
179 |
-
examples=example_images,
|
180 |
-
inputs=image_input,
|
181 |
-
outputs=result_output,
|
182 |
-
fn=classify_image,
|
183 |
-
cache_examples=True,
|
184 |
-
label="✨ Click an Example to Try!"
|
185 |
-
)
|
186 |
-
|
187 |
-
# Footer / Article section
|
188 |
-
gr.Markdown(f"""
|
189 |
-
---
|
190 |
-
**How it Works:**
|
191 |
-
This application uses a fine-tuned [SigLIP](https://huggingface.co/docs/transformers/model_doc/siglip) vision model
|
192 |
-
specifically trained to differentiate between images generated by Artificial Intelligence and those created by humans.
|
193 |
-
|
194 |
-
**Model:**
|
195 |
-
* You can find the model card here: <a href='https://huggingface.co/{MODEL_IDENTIFIER}' target='_blank'>{MODEL_IDENTIFIER}</a>
|
196 |
-
|
197 |
-
**Training Code:**
|
198 |
-
Fine tuning code available at [https://exnrt.com/blog/ai/fine-tuning-siglip2/](https://exnrt.com/blog/ai/fine-tuning-siglip2/).
|
199 |
-
""",
|
200 |
-
elem_id="app-footer"
|
201 |
-
)
|
202 |
-
|
203 |
-
# Connect events
|
204 |
-
submit_button.click(fn=classify_image, inputs=image_input, outputs=result_output, api_name="classify_image_button")
|
205 |
-
image_input.change(fn=classify_image, inputs=image_input, outputs=result_output, api_name="classify_image_change")
|
206 |
|
207 |
# --- Launch the App ---
|
208 |
if __name__ == "__main__":
|
|
|
10 |
DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
11 |
|
12 |
# --- Suppress specific warnings ---
|
13 |
+
# Suppress the specific PIL warning about potential decompression bombs
|
14 |
warnings.filterwarnings("ignore", message="Possibly corrupt EXIF data.")
|
15 |
+
# Suppress transformers warning about loading weights without specifying revision
|
16 |
warnings.filterwarnings("ignore", message=".*You are using the default legacy behaviour.*")
|
17 |
|
18 |
+
|
19 |
# --- Load Model and Processor (Load once at startup) ---
|
20 |
print(f"Using device: {DEVICE}")
|
21 |
print(f"Loading processor from: {MODEL_IDENTIFIER}")
|
|
|
28 |
print("Model and processor loaded successfully.")
|
29 |
except Exception as e:
|
30 |
print(f"FATAL: Error loading model or processor: {e}")
|
31 |
+
# If the model fails to load, we raise an exception to stop the app
|
32 |
raise gr.Error(f"Failed to load the model: {e}. Cannot start the application.") from e
|
33 |
|
34 |
# --- Prediction Function ---
|
35 |
def classify_image(image_pil):
|
36 |
+
"""
|
37 |
+
Classifies an image as AI-generated or Human-made.
|
38 |
+
Args:
|
39 |
+
image_pil (PIL.Image.Image): Input image in PIL format.
|
40 |
+
Returns:
|
41 |
+
dict: A dictionary mapping class labels ('ai', 'human') to their
|
42 |
+
confidence scores. Returns an empty dict if input is None.
|
43 |
+
"""
|
44 |
if image_pil is None:
|
45 |
+
# Handle case where the user clears the image input
|
46 |
print("Warning: No image provided.")
|
47 |
+
return {} # Return empty dict, Gradio Label handles this
|
48 |
|
49 |
print("Processing image...")
|
50 |
try:
|
51 |
+
# Ensure image is RGB
|
52 |
image = image_pil.convert("RGB")
|
53 |
+
|
54 |
+
# Preprocess using the loaded processor
|
55 |
inputs = processor(images=image, return_tensors="pt").to(DEVICE)
|
56 |
|
57 |
+
# Perform inference
|
58 |
print("Running inference...")
|
59 |
with torch.no_grad():
|
60 |
outputs = model(**inputs)
|
61 |
logits = outputs.logits
|
62 |
|
63 |
+
# Get probabilities using softmax
|
64 |
+
# outputs.logits is shape [1, num_labels], softmax over the last dim
|
65 |
+
probabilities = torch.softmax(logits, dim=-1)[0] # Get probabilities for the first (and only) image
|
66 |
+
|
67 |
+
# Create a dictionary of label -> score
|
68 |
results = {}
|
69 |
for i, prob in enumerate(probabilities):
|
70 |
label = model.config.id2label[i]
|
71 |
+
results[label] = prob.item() # Use .item() to get Python float
|
72 |
|
73 |
print(f"Prediction results: {results}")
|
74 |
return results
|
75 |
+
|
76 |
except Exception as e:
|
77 |
print(f"Error during prediction: {e}")
|
78 |
+
# Optionally raise a Gradio error to show it in the UI
|
79 |
+
# raise gr.Error(f"Error processing image: {e}")
|
80 |
+
return {"Error": f"Processing failed: {e}"} # Or return an error message
|
81 |
|
82 |
+
# --- Gradio Interface Definition ---
|
83 |
+
|
84 |
+
# Define Example Images (Optional, but recommended)
|
85 |
+
# Create an 'examples' folder in your Space repo and put images there
|
86 |
example_dir = "examples"
|
87 |
example_images = []
|
88 |
+
if os.path.exists(example_dir):
|
89 |
for img_name in os.listdir(example_dir):
|
90 |
if img_name.lower().endswith(('.png', '.jpg', '.jpeg', '.webp')):
|
91 |
+
example_images.append(os.path.join(example_dir, img_name))
|
92 |
+
print(f"Found examples: {example_images}")
|
|
|
|
|
|
|
93 |
else:
|
94 |
+
print("No 'examples' directory found. Examples will not be shown.")
|
95 |
+
|
96 |
+
|
97 |
+
# Define the Gradio interface
|
98 |
+
iface = gr.Interface(
|
99 |
+
fn=classify_image,
|
100 |
+
inputs=gr.Image(type="pil", label="Upload Image", sources=["upload", "webcam", "clipboard"]), # Use PIL format as input
|
101 |
+
outputs=gr.Label(num_top_classes=2, label="Prediction Results"), # Use gr.Label for classification output
|
102 |
+
title="AI vs Human Image Detector",
|
103 |
+
description=(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
104 |
f"Upload an image to classify if it was likely generated by AI or created by a human. "
|
105 |
+
f"Uses the `{MODEL_IDENTIFIER}` model on Hugging Face. Running on **{str(DEVICE).upper()}**."
|
106 |
+
),
|
107 |
+
article=(
|
108 |
+
"<div>"
|
109 |
+
"<p>This tool uses a SigLIP model fine-tuned for distinguishing between AI-generated and human-made images.</p>"
|
110 |
+
f"<p>Model Card: <a href='https://huggingface.co/{MODEL_IDENTIFIER}' target='_blank'>{MODEL_IDENTIFIER}</a></p>"
|
111 |
+
Fine tuning code available at [https://exnrt.com/blog/ai/fine-tuning-siglip2/](https://exnrt.com/blog/ai/fine-tuning-siglip2/).
|
112 |
+
"</div>"
|
113 |
+
),
|
114 |
+
examples=example_images if example_images else None, # Only add examples if found
|
115 |
+
cache_examples= True if example_images else False, # Cache results for examples if they exist
|
116 |
+
allow_flagging="never" # Or "auto" if you want users to flag issues
|
117 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
118 |
|
119 |
# --- Launch the App ---
|
120 |
if __name__ == "__main__":
|