change requirements and model

app.py

@@ -186,7 +186,7 @@ def predict_emotion(audio_file, text_input, video_file):
     """Main prediction function"""
     
     if audio_file is None or video_file is None or not text_input.strip():
-        return "Please provide all three inputs: audio, text, and video", None
+        return "Please provide all three inputs: audio, text, and video", None, None, None, None
     
     try:
         # Preprocess
@@ -202,19 +202,25 @@ def predict_emotion(audio_file, text_input, video_file):
         
         # Get probabilities
         fused_probs = torch.softmax(fused_logits, dim=1)[0].cpu().numpy()
+        audio_probs = torch.softmax(a_logits, dim=1)[0].cpu().numpy()
+        text_probs = torch.softmax(t_logits, dim=1)[0].cpu().numpy()
+        video_probs = torch.softmax(v_logits, dim=1)[0].cpu().numpy()
         
         # Format results
-        result = {LABELS[i]: float(fused_probs[i]) for i in range(len(LABELS))}
+        fused_result = {LABELS[i]: float(fused_probs[i]) for i in range(len(LABELS))}
+        audio_result = {LABELS[i]: float(audio_probs[i]) for i in range(len(LABELS))}
+        text_result = {LABELS[i]: float(text_probs[i]) for i in range(len(LABELS))}
+        video_result = {LABELS[i]: float(video_probs[i]) for i in range(len(LABELS))}
         
         predicted_emotion = LABELS[fused_probs.argmax()]
         confidence = float(fused_probs.max())
         
         result_text = f"🎯 **Predicted Emotion: {predicted_emotion.upper()}**\n\n**Confidence: {confidence:.2%}**"
         
-        return result_text, result
+        return result_text, fused_result, audio_result, text_result, video_result
         
     except Exception as e:
-        return f"Error: {str(e)}", None
+        return f"Error: {str(e)}", None, None, None, None
 
 # Gradio Interface
 with gr.Blocks(title="Multimodal Emotion Recognition", theme=gr.themes.Soft()) as demo:
@@ -247,14 +253,18 @@ with gr.Blocks(title="Multimodal Emotion Recognition", theme=gr.themes.Soft()) a
         
         with gr.Column():
             result_text = gr.Markdown(label="Result")
-            result_output = gr.Label(label="📊 Prediction Results", num_top_classes=4)
+            
+            with gr.Accordion("📊 Detailed Predictions", open=True):
+                fused_output = gr.Label(label="🔗 Fused Prediction", num_top_classes=4)
+                audio_output = gr.Label(label="🎤 Audio-only Prediction", num_top_classes=4)
+                text_output = gr.Label(label="📝 Text-only Prediction", num_top_classes=4)
+                video_output = gr.Label(label="🎥 Video-only Prediction", num_top_classes=4)
     
     predict_btn.click(
         fn=predict_emotion,
         inputs=[audio_input, text_input, video_input],
-        outputs=[result_text, result_output]
+        outputs=[result_text, fused_output, audio_output, text_output, video_output]
     )
-
     
     gr.Markdown(
         """