File size: 6,911 Bytes
547247c
3870592
 
547247c
 
 
 
 
 
 
3870592
547247c
 
3870592
547247c
3870592
 
 
547247c
 
 
 
 
 
 
 
 
 
3870592
 
 
 
547247c
3870592
 
 
547247c
 
3870592
 
 
1a69472
 
c38d472
3870592
 
 
 
547247c
3870592
 
547247c
3870592
547247c
3870592
 
 
 
547247c
 
 
3870592
 
 
c38d472
547247c
 
3870592
 
547247c
 
 
3870592
 
 
547247c
 
3870592
 
547247c
 
 
3870592
 
 
 
 
547247c
3870592
547247c
3870592
547247c
3870592
b8dcf5d
 
3870592
 
 
 
 
 
 
547247c
3870592
 
 
547247c
3870592
 
 
 
 
 
 
 
 
 
c38d472
3870592
 
 
70b84aa
3870592
 
 
 
 
70b84aa
 
3870592
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
70b84aa
3870592
 
 
 
 
 
 
 
1a69472
3870592
 
 
 
 
 
 
c38d472
3870592
 
 
e7616c0
3870592
 
1a69472
3870592
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
547247c
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
"""
Document Forgery Detection – Professional Gradio Dashboard
Hugging Face Spaces Deployment
"""

import gradio as gr
import torch
import cv2
import numpy as np
from PIL import Image
import plotly.graph_objects as go
from pathlib import Path
import sys
import json

# -------------------------------------------------
# PATH SETUP
# -------------------------------------------------
sys.path.insert(0, str(Path(__file__).parent))

from src.models import get_model
from src.config import get_config
from src.data.preprocessing import DocumentPreprocessor
from src.data.augmentation import DatasetAwareAugmentation
from src.features.region_extraction import get_mask_refiner, get_region_extractor
from src.features.feature_extraction import get_feature_extractor
from src.training.classifier import ForgeryClassifier

# -------------------------------------------------
# CONSTANTS
# -------------------------------------------------
CLASS_NAMES = {0: "Copy-Move", 1: "Splicing", 2: "Generation"}
CLASS_COLORS = {
    0: (255, 0, 0),
    1: (0, 255, 0),
    2: (0, 0, 255),
}

# -------------------------------------------------
# FORGERY DETECTOR (UNCHANGED CORE LOGIC)
# -------------------------------------------------
class ForgeryDetector:
    def __init__(self):
        print("Loading models...")

        self.config = get_config("config.yaml")
        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

        self.model = get_model(self.config).to(self.device)
        checkpoint = torch.load("models/best_doctamper.pth", map_location=self.device)
        self.model.load_state_dict(checkpoint["model_state_dict"])
        self.model.eval()

        self.classifier = ForgeryClassifier(self.config)
        self.classifier.load("models/classifier")

        self.preprocessor = DocumentPreprocessor(self.config, "doctamper")
        self.augmentation = DatasetAwareAugmentation(self.config, "doctamper", is_training=False)
        self.mask_refiner = get_mask_refiner(self.config)
        self.region_extractor = get_region_extractor(self.config)
        self.feature_extractor = get_feature_extractor(self.config, is_text_document=True)

        print("βœ“ Models loaded")

    def detect(self, image):
        if isinstance(image, Image.Image):
            image = np.array(image)

        if image.ndim == 2:
            image = cv2.cvtColor(image, cv2.COLOR_GRAY2RGB)
        elif image.shape[2] == 4:
            image = cv2.cvtColor(image, cv2.COLOR_RGBA2RGB)

        original = image.copy()

        preprocessed, _ = self.preprocessor(image, None)
        augmented = self.augmentation(preprocessed, None)
        image_tensor = augmented["image"].unsqueeze(0).to(self.device)

        with torch.no_grad():
            logits, decoder_features = self.model(image_tensor)
            prob_map = torch.sigmoid(logits).cpu().numpy()[0, 0]

        binary = (prob_map > 0.5).astype(np.uint8)
        refined = self.mask_refiner.refine(binary, original_size=original.shape[:2])
        regions = self.region_extractor.extract(refined, prob_map, original)

        results = []
        for r in regions:
            features = self.feature_extractor.extract(
                preprocessed, r["region_mask"], [f.cpu() for f in decoder_features]
            )

            if features.ndim == 1:
                features = features.reshape(1, -1)

            if features.shape[1] != 526:
                pad = max(0, 526 - features.shape[1])
                features = np.pad(features, ((0, 0), (0, pad)))[:, :526]

            pred, conf = self.classifier.predict(features)
            if conf[0] > 0.6:
                results.append({
                    "bounding_box": r["bounding_box"],
                    "forgery_type": CLASS_NAMES[int(pred[0])],
                    "confidence": float(conf[0]),
                })

        overlay = self._draw_overlay(original, results)

        return overlay, {
            "num_detections": len(results),
            "detections": results,
        }

    def _draw_overlay(self, image, results):
        out = image.copy()
        for r in results:
            x, y, w, h = r["bounding_box"]
            fid = [k for k, v in CLASS_NAMES.items() if v == r["forgery_type"]][0]
            color = CLASS_COLORS[fid]

            cv2.rectangle(out, (x, y), (x + w, y + h), color, 2)
            label = f"{r['forgery_type']} ({r['confidence']*100:.1f}%)"
            cv2.putText(out, label, (x, y - 6),
                        cv2.FONT_HERSHEY_SIMPLEX, 0.5, color, 2)
        return out


detector = ForgeryDetector()

# -------------------------------------------------
# METRIC VISUALS
# -------------------------------------------------
def gauge(value, title):
    fig = go.Figure(go.Indicator(
        mode="gauge+number",
        value=value,
        title={"text": title},
        gauge={"axis": {"range": [0, 100]}, "bar": {"color": "#2563eb"}}
    ))
    fig.update_layout(height=240, margin=dict(t=40, b=20))
    return fig

# -------------------------------------------------
# GRADIO CALLBACK
# -------------------------------------------------
def run_detection(file):
    image = Image.open(file.name)
    overlay, result = detector.detect(image)

    avg_conf = (
        sum(d["confidence"] for d in result["detections"]) / max(1, result["num_detections"])
    ) * 100

    return (
        overlay,
        result,
        gauge(75, "Localization Dice (%)"),
        gauge(92, "Classifier Accuracy (%)"),
        gauge(avg_conf, "Avg Detection Confidence (%)"),
    )

# -------------------------------------------------
# UI
# -------------------------------------------------
with gr.Blocks(theme=gr.themes.Soft(), title="Document Forgery Detection") as demo:

    gr.Markdown("# πŸ“„ Document Forgery Detection System")

    with gr.Row():
        file_input = gr.File(label="Upload Document (Image/PDF)")
        detect_btn = gr.Button("Run Detection", variant="primary")

    output_img = gr.Image(label="Forgery Localization Result", type="numpy")

    with gr.Tabs():
        with gr.Tab("πŸ“Š Metrics"):
            with gr.Row():
                dice_plot = gr.Plot()
                acc_plot = gr.Plot()
                conf_plot = gr.Plot()

        with gr.Tab("🧾 Details"):
            json_out = gr.JSON()

        with gr.Tab("πŸ‘₯ Team"):
            gr.Markdown("""
            **Document Forgery Detection Project**

            - Krishnanandhaa β€” Model & Training  
            - Teammate 1 β€” Feature Engineering  
            - Teammate 2 β€” Evaluation  
            - Teammate 3 β€” Deployment  

            *Collaborators are added via Hugging Face Space settings.*
            """)

    detect_btn.click(
        run_detection,
        inputs=file_input,
        outputs=[output_img, json_out, dice_plot, acc_plot, conf_plot]
    )

if __name__ == "__main__":
    demo.launch()