import os
import random
from dataclasses import dataclass
from typing import List, Optional
import gradio as gr
import numpy as np
import torch
from PIL import Image
from skimage.feature import graycomatrix, graycoprops
from torchvision import transforms
NUM_ROUNDS = 10
PROB_THRESHOLD = 0.3
# Load the model
model = torch.jit.load("SuSy.pt")
@dataclass
class GameResults:
user_guess: str
model_guess: str
correct_answer: str
class GameState:
def __init__(self):
self.reset()
def reset(self):
self.user_score = 0
self.model_score = 0
self.current_round = 0
self.total_rounds = NUM_ROUNDS
self.game_images: List[str] = []
self.is_game_active = False
self.last_results: Optional[GameResults] = None
self.processing_submission = False
def start_new_game(self) -> bool:
if self.is_game_active:
return False
self.reset()
self.game_images = load_images()
self.is_game_active = True
return True
def can_submit_guess(self) -> bool:
return (
self.is_game_active and
not self.processing_submission and
self.current_round < self.total_rounds
)
def start_submission(self) -> bool:
if not self.can_submit_guess():
return False
self.processing_submission = True
return True
def finish_submission(self, results: GameResults):
if results.user_guess == results.correct_answer:
self.user_score += 1
if results.model_guess == results.correct_answer:
self.model_score += 1
self.last_results = results
self.current_round += 1
self.processing_submission = False
if self.current_round >= self.total_rounds:
self.is_game_active = False
def get_current_image(self) -> Optional[str]:
if not self.is_game_active or self.current_round >= len(self.game_images):
return None
return self.game_images[self.current_round]
def get_game_over_message(self) -> str:
if self.user_score > self.model_score:
return """
🎉 Congratulations! You won! 🎉
You've outperformed SuSy in detecting AI-generated images.
Click 'Start New Game' to play again.
"""
elif self.user_score < self.model_score:
return """
Better luck next time! SuSy won this round.
Keep practicing to improve your detection skills.
Click 'Start New Game' to try again.
"""
else:
return """
It's a tie! You matched SuSy's performance!
You're getting good at this.
Click 'Start New Game' to play again.
"""
def create_score_html(game_state: GameState):
results_html = ""
if game_state.last_results:
results_html = f"""
Last Round Results:
Your guess: {game_state.last_results.user_guess}
Model's guess: {game_state.last_results.model_guess}
Correct answer: {game_state.last_results.correct_answer}
"""
current_display_round = min(game_state.current_round + 1, game_state.total_rounds)
return f"""
Score Board
You
{game_state.user_score}
AI Model
{game_state.model_score}
Round: {current_display_round}/{game_state.total_rounds}
{results_html}
"""
def process_image(image):
# Set Parameters
top_k_patches = 5
patch_size = 224
# Get the image dimensions
width, height = image.size
# Calculate the number of patches
num_patches_x = width // patch_size
num_patches_y = height // patch_size
# Divide the image in patches
patches = np.zeros((num_patches_x * num_patches_y, patch_size, patch_size, 3), dtype=np.uint8)
for i in range(num_patches_x):
for j in range(num_patches_y):
x = i * patch_size
y = j * patch_size
patch = image.crop((x, y, x + patch_size, y + patch_size))
patches[i * num_patches_y + j] = np.array(patch)
# Compute the most relevant patches
dissimilarity_scores = []
for patch in patches:
transform_patch = transforms.Compose([transforms.PILToTensor(), transforms.Grayscale()])
grayscale_patch = transform_patch(Image.fromarray(patch)).squeeze(0)
glcm = graycomatrix(grayscale_patch, [5], [0], 256, symmetric=True, normed=True)
dissimilarity_scores.append(graycoprops(glcm, "contrast")[0, 0])
# Sort patch indices by their dissimilarity score
sorted_indices = np.argsort(dissimilarity_scores)[::-1]
# Extract top k patches and convert them to tensor
top_patches = patches[sorted_indices[:top_k_patches]]
top_patches = torch.from_numpy(np.transpose(top_patches, (0, 3, 1, 2))) / 255.0
# Predict patches
model.eval()
with torch.no_grad():
preds = model(top_patches)
# Process results
classes = ['Authentic', 'DALL·E 3', 'Stable Diffusion 1.x', 'MJ V5/V6', 'MJ V1/V2', 'Stable Diffusion XL']
mean_probs = preds.mean(dim=0).numpy()
# Create a dictionary of class probabilities
class_probs = {cls: prob for cls, prob in zip(classes, mean_probs)}
# Sort probabilities in descending order
sorted_probs = dict(sorted(class_probs.items(), key=lambda item: item[1], reverse=True))
return sorted_probs
def load_images() -> List[str]:
real_image_folder = "real_images"
fake_image_folder = "fake_images"
real_images = [os.path.join(real_image_folder, img) for img in os.listdir(real_image_folder)]
fake_images = [os.path.join(fake_image_folder, img) for img in os.listdir(fake_image_folder)]
selected_images = random.sample(real_images, NUM_ROUNDS // 2) + random.sample(fake_images, NUM_ROUNDS // 2)
random.shuffle(selected_images)
return selected_images
def start_game(state: Optional[GameState]):
# Initialize new game state if none exists
if state is None:
state = GameState()
if not state.start_new_game():
return [state] + [gr.update()] * 6
current_image = Image.open(state.get_current_image())
return [
state,
gr.update(value=current_image, visible=True),
gr.update(visible=False),
gr.update(visible=True, interactive=True),
gr.update(visible=True, interactive=True),
create_score_html(state),
gr.update(visible=False),
]
def submit_guess(user_guess: str, state: GameState):
if not state.can_submit_guess():
return [state] + [gr.update()] * 6
if not state.start_submission():
return [state] + [gr.update()] * 6
try:
current_image_path = state.get_current_image()
if not current_image_path:
return [state] + [gr.update()] * 6
current_image = Image.open(current_image_path)
model_prediction = process_image(current_image)
model_guess = "Real" if model_prediction['Authentic'] > PROB_THRESHOLD else "Fake"
correct_answer = "Real" if "real_images" in current_image_path else "Fake"
# Update game state with results
results = GameResults(user_guess, model_guess, correct_answer)
state.finish_submission(results)
# Check if game is over
if not state.is_game_active:
return [
state,
gr.update(value=None, visible=False),
gr.update(visible=True),
gr.update(visible=False),
gr.update(visible=False),
create_score_html(state),
gr.update(visible=True, value=state.get_game_over_message())
]
# Get next image
next_image_path = state.get_current_image()
if not next_image_path:
return [state] + [gr.update()] * 6
next_image = Image.open(next_image_path)
return [
state,
gr.update(value=next_image, visible=True),
gr.update(visible=False),
gr.update(visible=True, interactive=True),
gr.update(visible=True, interactive=True),
create_score_html(state),
gr.update(visible=False)
]
except Exception as e:
state.processing_submission = False
raise e
# Custom CSS
custom_css = """
#game-container {
max-width: 1200px;
margin: 0 auto;
padding: 20px;
}
#start-button {
max-width: 200px;
margin: 0 auto;
}
#guess-buttons {
display: flex;
gap: 10px;
justify-content: center;
margin-top: 20px;
}
.guess-button {
min-width: 120px;
}
.image-container img {
max-height: 640px !important;
width: auto !important;
object-fit: contain !important;
}
"""
# Define Gradio interface
with gr.Blocks(css=custom_css) as iface:
# State variable for the game
state = gr.State(None)
with gr.Column(elem_id="game-container"):
gr.HTML("""
""")
with gr.Row():
with gr.Column(scale=2):
image_display = gr.Image(
type="pil",
label="Current Image",
interactive=False,
visible=False,
elem_classes=["image-container"]
)
with gr.Row(elem_id="guess-buttons"):
real_button = gr.Button(
"Real",
visible=False,
variant="primary",
elem_classes=["guess-button"]
)
fake_button = gr.Button(
"Fake",
visible=False,
variant="primary",
elem_classes=["guess-button"]
)
with gr.Column(scale=1):
score_display = gr.HTML()
with gr.Row():
with gr.Column(elem_id="start-button"):
start_button = gr.Button("Start New Game", variant="primary", size="sm")
feedback_display = gr.Markdown(visible=False)
# Event handlers
start_button.click(
fn=start_game,
inputs=[state],
outputs=[
state,
image_display,
start_button,
real_button,
fake_button,
score_display,
feedback_display
]
)
real_button.click(
fn=lambda state: submit_guess("Real", state),
inputs=[state],
outputs=[
state,
image_display,
start_button,
real_button,
fake_button,
score_display,
feedback_display,
],
)
fake_button.click(
fn=lambda state: submit_guess("Fake", state),
inputs=[state],
outputs=[
state,
image_display,
start_button,
real_button,
fake_button,
score_display,
feedback_display,
],
)
# Launch the interface
iface.launch()
