Update app.py

app.py

@@ -1,87 +1,128 @@
 import gradio as gr
-from transformers import pipeline
-from PIL import Image
+from transformers import AutoTokenizer, AutoModelForSequenceClassification, AutoImageProcessor, AutoModelForImageClassification
+from torch.nn.functional import sigmoid
 import torch
-from torchvision import transforms
-from torchvision.models import resnet50
-import torch.nn.functional as F
-
-# Load text emotion detection pipeline
-text_emotion_pipeline = pipeline("text-classification", model="SamLowe/roberta-base-go_emotions", top_k=3)
+from PIL import Image
 
-# Load image emotion detection model
-image_model = torch.hub.load('pytorch/vision', 'resnet50', pretrained=False)
-image_model.fc = torch.nn.Linear(image_model.fc.in_features, 7)
-image_model.load_state_dict(torch.hub.load_state_dict_from_url(
-    'https://huggingface.co/Celal11/resnet-50-finetuned-FER2013-0.001/resolve/main/pytorch_model.bin',
-    map_location=torch.device('cpu')
-))
-image_model.eval()
+# Load text emotion model
+model_name = "SamLowe/roberta-base-go_emotions"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForSequenceClassification.from_pretrained(model_name)
 
-image_emotions = ['Angry', 'Disgust', 'Fear', 'Happy', 'Sad', 'Surprise', 'Neutral']
+# Load image emotion model
+image_model_name = "Celal11/resnet-50-finetuned-FER2013-0.001"
+image_processor = AutoImageProcessor.from_pretrained(image_model_name)
+image_model = AutoModelForImageClassification.from_pretrained(image_model_name)
 
-transform = transforms.Compose([
-    transforms.Resize((224, 224)),
-    transforms.ToTensor()
-])
+# Analyze image emotion using processor and model
+def analyze_image_emotion(image):
+    if image is None:
+        return "No image provided."
+    inputs = image_processor(images=image, return_tensors="pt")
+    with torch.no_grad():
+        logits = image_model(**inputs).logits
+    probs = torch.nn.functional.softmax(logits, dim=1)[0]
+    pred_idx = torch.argmax(probs).item()
+    label = image_model.config.id2label[pred_idx]
+    score = probs[pred_idx].item()
+    return f"{label} ({score:.2f})"
 
-# Predefined reflections (simplified mapping)
-spiritual_reflections = {
-    "Sadness": "For indeed, with hardship [will be] ease. (Qur’an 94:6)",
-    "Joy": "Say, ‘In the bounty of Allah and in His mercy – in that let them rejoice.’ (Qur’an 10:58)",
-    "Fear": "And whoever fears Allah – He will make for him a way out. (Qur’an 65:2)",
-    "Anger": "Those who restrain anger and who pardon the people – Allah loves the doers of good. (Qur’an 3:134)",
-    "Surprise": "They plan, and Allah plans. Surely, Allah is the best of planners. (Qur’an 8:30)",
-    "Disgust": "Indeed, the most noble of you in the sight of Allah is the most righteous. (Qur’an 49:13)",
-    "Neutral": "Verily, in the remembrance of Allah do hearts find rest. (Qur’an 13:28)"
+# Emotion label to icon mapping (subset)
+emotion_icons = {
+    "admiration": "😍",
+    "amusement": "😅",
+    "anger": "😡",
+    "annoyance": "😑",
+    "approval": "👍",
+    "caring": "💗",
+    "confusion": "🤔",
+    "curiosity": "😮",
+    "desire": "🤤",
+    "disappointment": "😞",
+    "disapproval": "👎",
+    "disgust": "🤮",
+    "embarrassment": "😳",
+    "excitement": "🎉",
+    "fear": "😱",
+    "gratitude": "🙏",
+    "grief": "😭",
+    "joy": "😃",
+    "love": "❤️",
+    "nervousness": "🤧",
+    "optimism": "😊",
+    "pride": "😎",
+    "realization": "🤯",
+    "relief": "😌",
+    "remorse": "😔",
+    "sadness": "😢",
+    "surprise": "😲",
+    "neutral": "😐"
 }
 
-# Text input handler
-def analyze_text_emotion(text):
-    emotions = text_emotion_pipeline(text)
-    top_emotion = emotions[0]["label"]
-    reflection = spiritual_reflections.get(top_emotion, "Reflect with patience and prayer. (Qur’an 2:153)")
-    return top_emotion, reflection
-
-# Image input handler
-def analyze_image_emotion(image):
-    img_tensor = transform(image).unsqueeze(0)
+# Analyze text emotion
+def get_emotions(text, threshold):
+    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
     with torch.no_grad():
-        logits = image_model(img_tensor)
-        probs = F.softmax(logits, dim=1)[0]
-    top_idx = torch.argmax(probs).item()
-    top_emotion = image_emotions[top_idx]
-    reflection = spiritual_reflections.get(top_emotion, "Reflect with patience and prayer. (Qur’an 2:153)")
-    return top_emotion, reflection
+        logits = model(**inputs).logits
+    probs = sigmoid(logits)[0]
 
-# Gradio UI
-with gr.Blocks() as demo:
-    gr.Markdown("## 🧠 EmotionLens — AI-Powered Emotional Intelligence with Islamic Wisdom")
-    with gr.Tabs():
-        with gr.TabItem("Single Input"):
-            with gr.Row():
-                text_input = gr.Textbox(lines=2, placeholder="Enter text here...")
-                img_input = gr.Image(type="pil")
+    labels = [model.config.id2label[i] for i, p in enumerate(probs) if p > threshold]
+    icons = [emotion_icons.get(label, '') + ' ' + label.capitalize() + f" ({probs[i]:.2f})" for i, label in enumerate(labels)]
 
-            with gr.Row():
-                submit_btn = gr.Button("Submit")
+    return ", ".join(icons) if icons else "No strong emotion detected."
 
-            with gr.Column():
-                emotion_output = gr.Textbox(label="Emotion Detection")
-                reflection_output = gr.Textbox(label="Spiritual Reflection")
+# Combined analysis
+def analyze_combined(text, threshold, image):
+    text_result = get_emotions(text, threshold)
+    image_result = analyze_image_emotion(image)
+    return text_result, image_result
 
-            def combined_handler(text, img):
-                if text:
-                    return analyze_text_emotion(text)
-                elif img:
-                    return analyze_image_emotion(img)
-                else:
-                    return "No input", "Please provide text or image."
+# Gradio UI
+custom_css = """
+body {
+  background: linear-gradient(to right, #f9f9f9, #d4ecff);
+  font-family: 'Segoe UI', sans-serif;
+}
+.gr-button {
+  background-color: #007BFF !important;
+  color: white !important;
+  border-radius: 8px !important;
+  font-weight: bold;
+}
+.gr-button:hover {
+  background-color: #0056b3 !important;
+}
+.gr-textbox {
+  border-radius: 8px !important;
+  border: 1px solid #ccc !important;
+  padding: 10px !important;
+}
+.output-textbox {
+  font-size: 1.5rem;
+  font-weight: bold;
+  color: #333;
+  background-color: #f1f9ff;
+  border-radius: 8px;
+  padding: 10px;
+  border: 1px solid #007BFF;
+}
+"""
 
-            submit_btn.click(
-                fn=combined_handler,
-                inputs=[text_input, img_input],
-                outputs=[emotion_output, reflection_output]
-            )
+demo = gr.Interface(
+    fn=analyze_combined,
+    inputs=[
+        gr.Textbox(lines=5, placeholder="Write a sentence or a full paragraph...", label="Your Text"),
+        gr.Slider(minimum=0.1, maximum=0.9, value=0.3, step=0.05, label="Threshold"),
+        gr.Image(type="pil", label="Upload Face Photo")
+    ],
+    outputs=[
+        gr.Textbox(label="Detected Text Emotions", elem_classes=["output-textbox"]),
+        gr.Textbox(label="Detected Photo Emotion", elem_classes=["output-textbox"])
+    ],
+    title="🥰 Multi-Modal Emotion Detector",
+    description="Analyze emotion from both text and a facial photo. Adjust the threshold for text emotion sensitivity.",
+    theme="default",
+    css=custom_css
+)
 
 demo.launch()