new app.py

app.py

@@ -1,3 +1,8 @@
+# =========================================================
+# PROFESSIONAL MULTIMODAL EMOTION RECOGNITION UI
+# Sky Blue + White Theme
+# =========================================================
+
 import gradio as gr
 import torch
 import torch.nn as nn
@@ -23,11 +28,18 @@ import subprocess
 # =========================================================
 
 DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
 SAMPLE_RATE = 16000
+
 TEXT_MAX_LEN = 64
+
 LABELS = ["angry", "happy", "neutral", "sad"]
 
-gpu_status = "🟢 GPU Enabled" if torch.cuda.is_available() else "🔴 CPU Mode"
+gpu_status = (
+    "GPU Enabled"
+    if torch.cuda.is_available()
+    else "Running on CPU"
+)
 
 # =========================================================
 # LOAD PROCESSORS
@@ -46,7 +58,9 @@ tokenizer = AutoTokenizer.from_pretrained(
 # =========================================================
 
 class ResNetVideoEncoder(nn.Module):
+
     def __init__(self, out_dim=768):
+
         super().__init__()
 
         base = models.resnet18(pretrained=False)
@@ -119,6 +133,7 @@ class HBF(nn.Module):
         self.drop = nn.Dropout(0.1)
 
         self.act1 = nn.GELU()
+
         self.act2 = nn.Tanh()
 
         self.n = n_layers
@@ -176,10 +191,6 @@ class AVVideoModel(nn.Module):
 
         self.fc = nn.Linear(768, num_classes)
 
-        self.fc_audio = nn.Linear(768, num_classes)
-        self.fc_text = nn.Linear(768, num_classes)
-        self.fc_video = nn.Linear(768, num_classes)
-
     def forward(
         self,
         audio,
@@ -213,17 +224,15 @@ class AVVideoModel(nn.Module):
 
         v_pool = self.v_enc(video)
 
-        a_pool = torch.nan_to_num(a_pool)
-        t_pool = torch.nan_to_num(t_pool)
-        v_pool = torch.nan_to_num(v_pool)
-
-        fused = self.hbf(a_pool, t_pool, v_pool)
-
-        fused = torch.nan_to_num(fused)
+        fused = self.hbf(
+            a_pool,
+            t_pool,
+            v_pool
+        )
 
-        fused_logits = self.fc(fused)
+        logits = self.fc(fused)
 
-        return fused_logits
+        return logits
 
 # =========================================================
 # LOAD MODEL
@@ -246,11 +255,11 @@ try:
 
     model.eval()
 
-    print("✅ Model loaded successfully")
+    print("Model loaded successfully")
 
 except Exception as e:
 
-    print(f"❌ Error loading model: {e}")
+    print(f"Model loading failed: {e}")
 
 # =========================================================
 # VIDEO PROCESSING
@@ -316,58 +325,42 @@ def extract_video_frames(
 
 def extract_audio_from_video(video_path):
 
-    try:
-
-        audio_path = tempfile.NamedTemporaryFile(
-            delete=False,
-            suffix=".wav"
-        ).name
-
-        command = [
-            "ffmpeg",
-            "-i", video_path,
-            "-vn",
-            "-acodec", "pcm_s16le",
-            "-ar", str(SAMPLE_RATE),
-            "-ac", "1",
-            "-y",
-            audio_path
-        ]
-
-        subprocess.run(
-            command,
-            stdout=subprocess.DEVNULL,
-            stderr=subprocess.DEVNULL,
-            check=True
-        )
-
-        return audio_path
-
-    except Exception as e:
+    audio_path = tempfile.NamedTemporaryFile(
+        delete=False,
+        suffix=".wav"
+    ).name
+
+    command = [
+        "ffmpeg",
+        "-i", video_path,
+        "-vn",
+        "-acodec", "pcm_s16le",
+        "-ar", str(SAMPLE_RATE),
+        "-ac", "1",
+        "-y",
+        audio_path
+    ]
+
+    subprocess.run(
+        command,
+        stdout=subprocess.DEVNULL,
+        stderr=subprocess.DEVNULL,
+        check=True
+    )
 
-        raise ValueError(
-            f"Audio extraction failed: {str(e)}"
-        )
+    return audio_path
 
 # =========================================================
-# TRANSCRIPTION
+# WHISPER
 # =========================================================
 
 whisper_model = whisper.load_model("base")
 
 def transcribe_audio(audio_path):
 
-    try:
-
-        result = whisper_model.transcribe(audio_path)
-
-        return result["text"].strip()
+    result = whisper_model.transcribe(audio_path)
 
-    except Exception as e:
-
-        raise ValueError(
-            f"Transcription failed: {str(e)}"
-        )
+    return result["text"].strip()
 
 # =========================================================
 # PREPROCESSING
@@ -379,7 +372,6 @@ def preprocess_inputs(
     video_path
 ):
 
-    # AUDIO
     wav, _ = librosa.load(
         audio_path,
         sr=SAMPLE_RATE
@@ -397,7 +389,6 @@ def preprocess_inputs(
         audio_values
     ).to(DEVICE)
 
-    # TEXT
     text_clean = re.sub(
         r"[^a-zA-Z0-9\s]",
         "",
@@ -416,14 +407,8 @@ def preprocess_inputs(
 
     text_mask = text_inputs.attention_mask.to(DEVICE)
 
-    # VIDEO
     frames = extract_video_frames(video_path)
 
-    if frames is None:
-        raise ValueError(
-            "Could not extract video frames"
-        )
-
     frames_tensor = (
         torch.tensor(frames)
         .permute(0, 3, 1, 2)
@@ -446,7 +431,7 @@ def preprocess_inputs(
     )
 
 # =========================================================
-# PREDICTION FUNCTION
+# PREDICTION
 # =========================================================
 
 def predict_emotion(video_file):
@@ -454,24 +439,21 @@ def predict_emotion(video_file):
     if video_file is None:
 
         return (
-            "❌ Please upload a video",
+            "Please upload a video.",
             None,
             ""
         )
 
     try:
 
-        # EXTRACT AUDIO
         audio_path = extract_audio_from_video(
             video_file
         )
 
-        # TRANSCRIBE
         transcribed_text = transcribe_audio(
             audio_path
         )
 
-        # PREPROCESS
         (
             audio,
             audio_mask,
@@ -484,7 +466,6 @@ def predict_emotion(video_file):
             video_file
         )
 
-        # INFERENCE
         with torch.no_grad():
 
             logits = model(
@@ -505,25 +486,20 @@ def predict_emotion(video_file):
             for i in range(len(LABELS))
         }
 
-        predicted_emotion = LABELS[probs.argmax()]
+        predicted_emotion = LABELS[
+            probs.argmax()
+        ]
 
         confidence = float(probs.max())
 
-        emoji_map = {
-            "happy": "😄",
-            "sad": "😢",
-            "angry": "😠",
-            "neutral": "😐"
-        }
-
         result_text = f"""
-# {emoji_map[predicted_emotion]} {predicted_emotion.upper()}
+        ## Predicted Emotion
+        
+        ### {predicted_emotion.upper()}
+        
+        Confidence Score: {confidence:.2%}
+        """
 
-## Confidence Score
-### {confidence:.2%}
-"""
-
-        # CLEANUP
         if os.path.exists(audio_path):
             os.remove(audio_path)
 
@@ -536,51 +512,75 @@ def predict_emotion(video_file):
     except Exception as e:
 
         return (
-            f"❌ Error: {str(e)}",
+            f"Error: {str(e)}",
             None,
             ""
         )
 
 # =========================================================
-# CUSTOM CSS
+# CLEAN PROFESSIONAL CSS
 # =========================================================
 
 custom_css = """
 
 body {
-    background: linear-gradient(
-        to right,
-        #0f172a,
-        #1e293b
-    );
+    background: #eaf6ff;
 }
 
 .gradio-container {
-    max-width: 1200px !important;
+    max-width: 1250px !important;
     margin: auto;
+    padding-top: 20px;
 }
 
 .main-title {
     text-align: center;
     font-size: 48px;
     font-weight: 800;
-    color: white;
-    margin-top: 20px;
+    color: #0f4c81;
+    margin-bottom: 10px;
 }
 
 .subtitle {
     text-align: center;
-    font-size: 18px;
-    color: #cbd5e1;
-    margin-bottom: 25px;
+    font-size: 20px;
+    color: #3b82b6;
+    margin-bottom: 30px;
+}
+
+.section-box {
+    background: white;
+    border-radius: 16px;
+    padding: 20px;
+    box-shadow: 0px 4px 12px rgba(0,0,0,0.08);
 }
 
 .footer {
     text-align: center;
-    color: #94a3b8;
+    color: #4b5563;
     margin-top: 30px;
     font-size: 14px;
 }
+
+.gr-button {
+    background: #38bdf8 !important;
+    border: none !important;
+    color: white !important;
+    font-weight: 600 !important;
+}
+
+.gr-button:hover {
+    background: #0ea5e9 !important;
+}
+
+h1, h2, h3, h4 {
+    color: #0f4c81 !important;
+}
+
+label {
+    color: #0f4c81 !important;
+    font-weight: 600 !important;
+}
 """
 
 # =========================================================
@@ -588,76 +588,89 @@ body {
 # =========================================================
 
 with gr.Blocks(
-    theme=gr.themes.Glass(),
-    css=custom_css,
-    title="Emotion Recognition"
+    title="Multimodal Emotion Recognition",
+    theme=gr.themes.Soft(),
+    css=custom_css
 ) as demo:
 
     # HEADER
+
     gr.HTML("""
     <div class="main-title">
-        🎭 Multimodal Emotion Recognition
+        Multimodal Emotion Recognition
     </div>
 
     <div class="subtitle">
-        AI-powered Emotion Detection using Audio, Text & Video Fusion
+        AI-based Emotion Detection using Audio, Text and Video Fusion
     </div>
     """)
 
-    gr.Markdown(f"### {gpu_status}")
+    # STATUS
+
+    gr.Markdown(
+        f"### System Status: {gpu_status}"
+    )
+
+    # INFO PANELS
 
-    # INFO SECTION
     with gr.Row():
 
         with gr.Column():
 
-            gr.Markdown("""
-            ### 🔍 Modalities
+            with gr.Group():
 
-            - 🎤 Audio
-            - 📝 Text
-            - 🎥 Video
-            """)
+                gr.Markdown("""
+                ### Modalities Used
+                
+                - Audio Analysis  
+                - Speech Transcription  
+                - Facial Expression Analysis  
+                """)
 
         with gr.Column():
 
-            gr.Markdown("""
-            ### 🤖 Models Used
+            with gr.Group():
 
-            - Wav2Vec2
-            - BERT
-            - ResNet18
-            - Whisper
-            """)
+                gr.Markdown("""
+                ### Models Used
+                
+                - Wav2Vec2  
+                - BERT  
+                - ResNet18  
+                - Whisper  
+                """)
 
     gr.Markdown("---")
 
     # MAIN SECTION
+
     with gr.Row(equal_height=True):
 
-        # LEFT SIDE
+        # LEFT
+
         with gr.Column(scale=1):
 
-            gr.Markdown("## 📤 Upload Video")
+            gr.Markdown("## Upload Video")
 
             video_input = gr.Video(
                 label="Input Video",
-                height=350
+                height=400
             )
 
             predict_btn = gr.Button(
-                "🚀 Analyze Emotion",
+                "Analyze Emotion",
                 variant="primary",
                 size="lg"
             )
 
-        # RIGHT SIDE
+        # RIGHT
+
         with gr.Column(scale=1):
 
-            gr.Markdown("## 📊 Results")
+            gr.Markdown("## Results")
 
             result_text = gr.Markdown(
-                value="Upload a video to begin analysis"
+                value="Upload a video and click Analyze Emotion."
             )
 
             result_output = gr.Label(
@@ -666,50 +679,46 @@ with gr.Blocks(
             )
 
             transcription_output = gr.Textbox(
-                label="📝 Transcribed Text",
-                lines=5,
+                label="Transcribed Text",
+                lines=6,
                 interactive=False
             )
 
     gr.Markdown("---")
 
-    # ABOUT MODEL
+    # ABOUT SECTION
+
     with gr.Accordion(
-        "ℹ️ About This Model",
+        "About the Model",
         open=False
     ):
 
         gr.Markdown("""
-        This system combines:
-
-        ### 🎤 Audio Analysis
-        Wav2Vec2 captures emotional tone and speech patterns.
-
-        ### 📝 Text Analysis
-        BERT analyzes semantic meaning from transcripts.
-
-        ### 🎥 Video Analysis
-        ResNet18 extracts facial expression features.
-
-        ### 🧠 Fusion Network
-        HBF combines all modalities for final prediction.
-
-        ---
-        Supported Emotions:
-        - Angry
-        - Happy
-        - Neutral
-        - Sad
+        This multimodal system combines:
+        
+        - Audio features using Wav2Vec2  
+        - Text understanding using BERT  
+        - Video feature extraction using ResNet18  
+        - Hybrid Fusion Block for final prediction
+        
+        Supported emotions:
+        
+        - Angry  
+        - Happy  
+        - Neutral  
+        - Sad  
         """)
 
     # FOOTER
+
     gr.HTML("""
     <div class="footer">
-        Built with ❤️ using PyTorch, Transformers, Whisper & Gradio
+        Built using PyTorch, Transformers, Whisper and Gradio
     </div>
     """)
 
     # BUTTON ACTION
+
     predict_btn.click(
         fn=predict_emotion,
         inputs=[video_input],