Update app.py

app.py

@@ -1,18 +1,24 @@
 import gradio as gr
 from transformers import pipeline
+import time
 
 # 1. Initialize your custom pipeline 
-# Replace with your exact Hugging Face model ID (e.g., "username/your-model")
 model_id = "google/vit-base-patch16-224" 
 classifier = pipeline("image-classification", model=model_id)
 
-# 2. Shared prediction function
+# 2. Optimized prediction function that measures latency
 def predict_image(image, min_confidence):
     if image is None:
-        return {}
+        return {}, "0 ms"
     
-    # Run the model
+    # Start the clock right before the model processes the frame
+    start_time = time.perf_counter()
     predictions = classifier(image)
+    end_time = time.perf_counter()
+    
+    # Calculate inference latency in milliseconds
+    latency_ms = int((end_time - start_time) * 1000)
+    latency_text = f"{latency_ms} ms"
     
     # Filter and format results based on the slider value
     filtered_results = {}
@@ -21,35 +27,43 @@ def predict_image(image, min_confidence):
         if score >= min_confidence:
             filtered_results[pred["label"]] = score
             
-    return filtered_results
+    return filtered_results, latency_text
 
-# 3. Custom Gradio Layout with Tabs
-with gr.Blocks(title="Advanced Image Classifier") as demo:
-    gr.Markdown("# 🖼️ Multi-Input Image Classification")
-    gr.Markdown("Choose your input method, adjust the threshold, and click classify.")
+# 3. Snapshot utility to save live images
+def save_snapshot(image):
+    if image is None:
+        return "No image captured to save."
+    
+    # Generate a unique timestamped file name
+    filename = f"snapshot_{int(time.time())}.jpg"
+    image.save(filename)
+    return f"✅ Snapshot successfully saved as '{filename}' inside Space directory!"
+
+# 4. Custom Layout Layout
+with gr.Blocks(title="Real-Time Analytics Classifier") as demo:
+    gr.Markdown("# ⚡ Real-Time Webcam Analytics")
+    gr.Markdown("Stream live video to measure pipeline inference speeds and save high-scoring frames.")
     
     with gr.Row():
         with gr.Column(scale=1):
             
             # --- TABBED INPUT SECTION ---
             with gr.Tabs():
-                with gr.TabItem("Upload File"):
-                    upload_input = gr.Image(type="pil", label="Upload Image")
-                    # We create separate submit buttons per tab for clear UX
-                    submit_upload = gr.Button("Classify Uploaded Image", variant="primary")
+                with gr.TabItem("Live Webcam Stream"):
+                    webcam_input = gr.Image(sources="webcam", type="pil", label="Live Stream Feed")
+                    
+                    # Performance tracking row added inside the tab view
+                    with gr.Row():
+                        latency_box = gr.Textbox(label="Model Inference Speed", value="0 ms", interactive=False)
+                        snapshot_btn = gr.Button("📸 Save Current Frame", variant="secondary")
                     
-                    # Examples only apply nicely to the upload field
-                    gr.Examples(
-                        examples=[["example_dog.jpg"], ["example_car.jpg"]],
-                        inputs=upload_input,
-                        label="Click an example to test"
-                    )
+                    snapshot_status = gr.Markdown("") # Status feedback text block
                     
-                with gr.TabItem("Use Webcam"):
-                    webcam_input = gr.Image(sources=["webcam"], type="pil", label="Take a Snapshot")
-                    submit_webcam = gr.Button("Classify Webcam Image", variant="primary")
+                with gr.TabItem("Upload File"):
+                    upload_input = gr.Image(type="pil", label="Static File Upload")
+                    submit_upload = gr.Button("Classify Uploaded Image", variant="primary")
 
-            # --- SHARED PARAMETERS ---
+            # --- SHARED CONFIGURATIONS ---
             threshold_slider = gr.Slider(
                 minimum=0.0, 
                 maximum=1.0, 
@@ -59,22 +73,35 @@ with gr.Blocks(title="Advanced Image Classifier") as demo:
             )
             
         with gr.Column(scale=1):
-            # --- OUTPUT SECTION ---
+            # --- REAL-TIME LABEL OUTPUT ---
             output_labels = gr.Label(num_top_classes=5, label="Predictions")
 
-    # 4. Wire up events for both input sources
-    submit_upload.click(
-        fn=predict_image, 
-        inputs=[upload_input, threshold_slider], 
-        outputs=output_labels
+    # 5. --- LIVE EVENT ROUTING ---
+    
+    # Live streaming updates predictions and latency counters concurrently
+    webcam_input.stream(
+        fn=predict_image,
+        inputs=[webcam_input, threshold_slider],
+        outputs=[output_labels, latency_box],
+        stream_every=0.2,
+        time_limit=300,
+        concurrency_limit=5
     )
     
-    submit_webcam.click(
+    # Saves a snapshot of whatever is currently on screen
+    snapshot_btn.click(
+        fn=save_snapshot,
+        inputs=[webcam_input],
+        outputs=[snapshot_status]
+    )
+    
+    # Manual submit routing for standard file uploads
+    submit_upload.click(
         fn=predict_image, 
-        inputs=[webcam_input, threshold_slider], 
-        outputs=output_labels
+        inputs=[upload_input, threshold_slider], 
+        outputs=[output_labels, latency_box]
     )
 
-# 5. Launch the application
+# 6. Launch the application
 if __name__ == "__main__":
     demo.launch()