Update app.py

app.py

@@ -219,7 +219,7 @@ import tempfile
 import os
 from PIL import Image
 import tensorflow as tf
-from transformers import pipeline
+from transformers import AutoModelForSequenceClassification, AutoTokenizer, pipeline
 from tensorflow.keras.applications import Xception, EfficientNetB7
 from tensorflow.keras.models import Model
 from tensorflow.keras.layers import Dense, GlobalAveragePooling2D
@@ -231,15 +231,30 @@ st.set_page_config(page_title="Fake & Deepfake Detection", layout="wide")
 st.title("\U0001F4F0 Fake News & Deepfake Detection Tool")
 st.write("\U0001F680 Detect Fake News, Deepfake Images, and Videos using AI")
 
-# Load Models
-fake_news_detector = pipeline("zero-shot-classification", model="facebook/bart-large-mnli")
+# Load Fake News Detection Model
+model_name = "mrm8488/bert-mini-fake-news"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+fake_news_model = AutoModelForSequenceClassification.from_pretrained(model_name)
+
+def detect_fake_news(text):
+    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
+    outputs = fake_news_model(**inputs)
+    probs = outputs.logits.softmax(dim=1).detach().numpy()[0]
+    
+    fake_prob = probs[0]
+    real_prob = probs[1]
+    
+    if fake_prob > real_prob:
+        return {"label": "FAKE", "confidence": fake_prob}
+    else:
+        return {"label": "REAL", "confidence": real_prob}
 
 # Load Deepfake Detection Models
 base_model_image = Xception(weights="imagenet", include_top=False)
-base_model_image.trainable = False  # Freeze base layers
+base_model_image.trainable = False
 x = GlobalAveragePooling2D()(base_model_image.output)
 x = Dense(1024, activation="relu")(x)
-x = Dense(1, activation="sigmoid")(x)  # Sigmoid for probability output
+x = Dense(1, activation="sigmoid")(x)
 deepfake_image_model = Model(inputs=base_model_image.input, outputs=x)
 
 base_model_video = EfficientNetB7(weights="imagenet", include_top=False)
@@ -251,10 +266,10 @@ deepfake_video_model = Model(inputs=base_model_video.input, outputs=x)
 
 # Function to Preprocess Image
 def preprocess_image(image_path):
-    img = load_img(image_path, target_size=(299, 299))  # Xception expects 299x299
+    img = load_img(image_path, target_size=(299, 299))
     img = img_to_array(img)
     img = np.expand_dims(img, axis=0)
-    img /= 255.0  # Normalize pixel values
+    img /= 255.0
     return img
 
 # Function to Detect Deepfake Image
@@ -271,17 +286,11 @@ news_input = st.text_area("Enter News Text:", placeholder="Type here...")
 
 if st.button("Check News"):
     st.write("\U0001F50D Processing...")
-
-    labels = ["FAKE", "REAL"]  # Corrected labels
-    prediction = fake_news_detector(news_input, candidate_labels=labels)
-
-    label = prediction["labels"][0]  # Highest confidence label
-    confidence = prediction["scores"][0]  # Confidence of highest label
-
-    if label == "FAKE":
-        st.error(f"⚠️ Result: This news is FAKE. (Confidence: {confidence:.2f})")
+    result = detect_fake_news(news_input)
+    if result["label"] == "FAKE":
+        st.error(f"⚠️ Result: This news is FAKE. (Confidence: {result['confidence']:.2f})")
     else:
-        st.success(f"✅ Result: This news is REAL. (Confidence: {confidence:.2f})")
+        st.success(f"✅ Result: This news is REAL. (Confidence: {result['confidence']:.2f})")
 
 # ---- Deepfake Image Detection Section ----
 st.subheader("\U0001F4F8 Deepfake Image Detection")
@@ -296,7 +305,6 @@ if uploaded_image is not None:
     if st.button("Analyze Image"):
         st.write("\U0001F50D Processing...")
         result = detect_deepfake_image(temp_file.name)
-        
         if result["label"] == "FAKE":
             st.error(f"⚠️ Result: This image is a Deepfake. (Confidence: {result['score']:.2f})")
         else:
@@ -335,10 +343,10 @@ if uploaded_video is not None:
     if st.button("Analyze Video"):
         st.write("\U0001F50D Processing...")
         result = detect_deepfake_video(temp_file.name)
-        
         if result["label"] == "FAKE":
             st.warning(f"⚠️ Result: This video contains Deepfake elements. (Confidence: {result['score']:.2f})")
         else:
             st.success(f"✅ Result: This video is Real. (Confidence: {1 - result['score']:.2f})")
 
 st.markdown("🔹 **Developed for Fake News & Deepfake Detection Hackathon**")
+