Changed issues due to pull

app.py

@@ -4,7 +4,6 @@ from transformers import RobertaTokenizer, ViTImageProcessor
 from PIL import Image
 import numpy as np
 import torch
-<<<<<<< HEAD
 import os
 import time
 import logging
@@ -92,46 +91,10 @@ def predict_news(text, image):
         
         # Process text input
         inputs = tokenizer.encode_plus(text, add_special_tokens = True, return_tensors='np', max_length=80, truncation=True, padding='max_length')
-=======
-from torchvision.transforms import v2
-import os
-import time
-
-tokenizer = BertTokenizer.from_pretrained("bert-base-uncased")
-model_path = "./multimodal_model_optimized.onnx"
-ort_session = ort.InferenceSession(model_path)
-
-transform = v2.Compose([
-    v2.Resize((256, 256)),
-    v2.ToImage(),
-    v2.ToDtype(torch.float32, scale=True),
-    v2.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
-])
-
-labels = ["Fake", "Real"]
-
-def predict_news(text, image):
-    if text is None or text.strip() == "":
-        return {labels[0]: 0.0, labels[1]: 0.0}, None, "Please enter some text to analyze."
-    
-    if image is None:
-        return {labels[0]: 0.0, labels[1]: 0.0}, None, "Please upload an image to analyze."
-    
-    try:
-        inputs = tokenizer.encode_plus(
-            text, 
-            add_special_tokens=True,
-            return_tensors='np',  
-            max_length=80,
-            truncation=True, 
-            padding='max_length'
-        )
->>>>>>> 585173c095c709c00b2ab290bb8d69553911f0d5
         
         input_ids = inputs['input_ids']
         attention_mask = inputs['attention_mask']
         
-<<<<<<< HEAD
         logger.info(f"Input IDs shape: {input_ids.shape}")
         logger.info(f"Attention mask shape: {attention_mask.shape}")
         
@@ -198,97 +161,17 @@ def predict_news(text, image):
     except Exception as e:
         logger.error(f"Error during analysis: {str(e)}", exc_info=True)
         return {labels[0]: 0.0, labels[1]: 0.0, labels[2]: 0.0}, None, f"Error during analysis: {str(e)}"
-=======
-        image_tensor = transform(image).numpy()  
-        
-        ort_inputs = {
-            "input_ids": input_ids,
-            "attention_mask": attention_mask,
-            "image": image_tensor.reshape(1, 3, 256, 256)  # Ensure correct shape
-        }
-        
-        start_time = time.time()
-        outputs = ort_session.run(None, ort_inputs)
-        inference_time = time.time() - start_time
-        
-        logits = outputs[0]
-        probs = softmax(logits)[0]
-        
-        pred_idx = int(np.argmax(probs))
-        confidence = float(probs[pred_idx])
-        
-        if pred_idx == 1:  # Real
-            color = "green"
-            message = f"This content appears to be **REAL** with {confidence:.1%} confidence."
-        else:  # Fake
-            color = "red"
-            message = f"This content appears to be **FAKE** with {confidence:.1%} confidence."
-            
-        analysis = f"""
-        <div style='text-align: center; padding: 10px; background-color: {color}15; border-radius: 5px; margin-top: 10px;'>
-            <span style='font-size: 18px; color: {color}; font-weight: bold;'>{message}</span>
-            <p>Inference time: {inference_time:.3f} seconds</p>
-        </div>
-        """
-        
-        result = {labels[0]: float(probs[0]), labels[1]: float(probs[1])}
-        
-        interpretation = image_with_prediction(image, labels[pred_idx], confidence)
-        
-        return result, interpretation, analysis
-        
-    except Exception as e:
-        return {labels[0]: 0.0, labels[1]: 0.0}, None, f"Error during analysis: {str(e)}"
-
-def softmax(x):
-    """Compute softmax values for each sets of scores in x."""
-    e_x = np.exp(x - np.max(x, axis=1, keepdims=True))
-    return e_x / e_x.sum(axis=1, keepdims=True)
-
-def image_with_prediction(img, label, confidence):
-    """Return the original image with an overlay showing the prediction"""
-    from PIL import Image, ImageDraw, ImageFont
-    import io
-    
-    img_copy = img.copy()
-    draw = ImageDraw.Draw(img_copy)
-    
-    width, height = img_copy.size
-    
-    overlay = Image.new('RGBA', (width, 40), (0, 0, 0, 150))
-    img_copy.paste(overlay, (0, height-40), overlay)
-    
-    text = f"{label}: {confidence:.1%}"
-    
-    try:
-        font = ImageFont.truetype("arial.ttf", 20)
-    except IOError:
-        font = ImageFont.load_default()
-    
-    text_width = draw.textlength(text, font=font)
-    text_position = ((width - text_width) // 2, height - 35)
-    draw.text(text_position, text, fill=(255, 255, 255), font=font)
-    
-    return img_copy
->>>>>>> 585173c095c709c00b2ab290bb8d69553911f0d5
 
 examples = [
     ["COVID-19 vaccine causes severe side effects in 80% of recipients", "https://images.unsplash.com/photo-1605289982774-9a6fef564df8?q=80&w=1000&auto=format&fit=crop"],
     ["Scientists discover new species of deep-sea fish", "https://images.unsplash.com/photo-1524704796725-9fc3044a58b2?q=80&w=1000&auto=format&fit=crop"],
 ]
-
-<<<<<<< HEAD
+    
 # Build Gradio interface
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
     gr.Markdown(
         """
-        # 📰 Fake News Detector (BERT + VIT)
-=======
-with gr.Blocks(theme=gr.themes.Soft()) as demo:
-    gr.Markdown(
-        """
-        # 📰 Fake News Detector (BERT + ResNet)
->>>>>>> 585173c095c709c00b2ab290bb8d69553911f0d5
+        # 📰 Fake News Detector (RoBERTa + ViT)
         
         This multimodal AI system analyzes both text and images to detect potentially fake news content.
         Upload an image and enter a news headline to see if the combination is likely to be real or fake news.
@@ -321,22 +204,12 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
     
     gr.Markdown(
         """
-        ### How it works
-        
         This system combines:
-<<<<<<< HEAD
         - **RoBERTa**: Analyzes the textual content
         - **ViT**: Processes the image data
         - **Multimodal Fusion**: Combines both signals to make a prediction
         
         The model was trained on the Fakeddit dataset containing real and fake news pairs with both text and images.
-=======
-        - **BERT**: Analyzes the textual content
-        - **ResNet**: Processes the image data
-        - **Multimodal Fusion**: Combines both signals to make a prediction
-        
-        The model was trained on a dataset of real and fake news pairs containing both text and images.
->>>>>>> 585173c095c709c00b2ab290bb8d69553911f0d5
         """
     )
     
@@ -347,8 +220,5 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
     )
 
 if __name__ == "__main__":
-<<<<<<< HEAD
     logger.info("Starting Gradio application")
-=======
->>>>>>> 585173c095c709c00b2ab290bb8d69553911f0d5
     demo.launch()