feat: Implement memory management and queue optimization

Major improvements:
- Add intelligent memory management for both input and output queues
- Implement frame dropping strategy to prevent memory overflow
- Set output queue limit to 50 results with FIFO cleanup
- Add input queue management with real-time frame dropping
- Filter async results from callbacks and display to reduce noise
- Improve system stability and prevent queue-related hangs
- Add comprehensive logging for dropped frames and results

Performance enhancements:
- Maintain real-time processing by prioritizing latest frames
- Prevent memory accumulation that previously caused system freezes
- Ensure consistent queue size reporting and FPS calculations

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

cluster4npu_ui/core/functions/InferencePipeline.py

@@ -526,11 +526,24 @@ class InferencePipeline:
                         current_data.metadata['start_timestamp']
                     )
                     
+                    # Only put valid inference results into output queue
+                    if has_valid_inference:
+                        # Manage output queue size - maintain fixed upper limit for memory management
+                        MAX_OUTPUT_QUEUE_SIZE = 50  # Set maximum output queue size
+                        
+                        # If queue is getting full, remove old results to make space
+                        while self.pipeline_output_queue.qsize() >= MAX_OUTPUT_QUEUE_SIZE:
+                            try:
+                                dropped_result = self.pipeline_output_queue.get_nowait()
+                                # Track dropped results for debugging
+                                if not hasattr(self, '_dropped_results_count'):
+                                    self._dropped_results_count = 0
+                                self._dropped_results_count += 1
+                            except queue.Empty:
+                                break
+                        
                         try:
                             self.pipeline_output_queue.put(current_data, block=False)
-                        
-                        # Only count completed results if they contain valid inference
-                        if has_valid_inference:
                             self.completed_counter += 1
                             # Record output timestamp for FPS calculation
                             self._record_output_timestamp()
@@ -539,21 +552,17 @@ class InferencePipeline:
                             if self.completed_counter % 10 == 0:
                                 print(f"[{self.pipeline_name}] Processed {self.completed_counter} results")
                                 print(f"[{self.pipeline_name}] Queue sizes - Input: {self.pipeline_input_queue.qsize()}, Output: {self.pipeline_output_queue.qsize()}")
+                                # Show dropped results info if any
+                                if hasattr(self, '_dropped_results_count') and self._dropped_results_count > 0:
+                                    print(f"[{self.pipeline_name}] Dropped {self._dropped_results_count} old results for memory management")
                             
-                        # Call result callback
+                            # Call result callback for valid inference results
                             if self.result_callback:
                                 self.result_callback(current_data)
                                 
                         except queue.Full:
-                        # Drop oldest and add new
+                            # Fallback: should rarely happen due to pre-emptive cleaning above
-                        try:
+                            print(f"[{self.pipeline_name}] Warning: Output queue still full after cleanup")
-                            self.pipeline_output_queue.get_nowait()
-                            self.pipeline_output_queue.put(current_data, block=False)
-                            # Only record timestamp and count if valid inference result
-                            if has_valid_inference:
-                                self._record_output_timestamp()
-                        except queue.Empty:
-                            pass
                 else:
                     self.error_counter += 1
                     if self.error_callback:
@@ -567,10 +576,31 @@ class InferencePipeline:
         print(f"[{self.pipeline_name}] Coordinator stopped")
     
     def put_data(self, data: Any, timeout: float = 1.0) -> bool:
-        """Put data into pipeline"""
+        """Put data into pipeline with memory management"""
         try:
             self.pipeline_input_queue.put(data, timeout=timeout)
             return True
+        except queue.Full:
+            # Drop oldest frames to make space for new ones (for real-time processing)
+            try:
+                dropped_data = self.pipeline_input_queue.get_nowait()
+                self.pipeline_input_queue.put(data, block=False)
+                
+                # Track dropped frames for debugging
+                if not hasattr(self, '_dropped_frames_count'):
+                    self._dropped_frames_count = 0
+                self._dropped_frames_count += 1
+                
+                # Log occasionally to show frame dropping (every 50 drops)
+                if self._dropped_frames_count % 50 == 0:
+                    print(f"[{self.pipeline_name}] Dropped {self._dropped_frames_count} input frames for real-time processing")
+                
+                return True
+            except queue.Empty:
+                # Rare case: queue became empty between full check and get
+                try:
+                    self.pipeline_input_queue.put(data, block=False)
+                    return True
                 except queue.Full:
                     return False
     

cluster4npu_ui/core/functions/workflow_orchestrator.py

@@ -181,7 +181,7 @@ class WorkflowOrchestrator:
                 self.result_callback(result_dict)
                 
         except Exception as e:
-            print(f"❌ Error handling result: {e}")
+            print(f"Error handling result: {e}")
 
     def _parse_resolution(self, resolution_str: Optional[str]) -> Optional[tuple[int, int]]:
         """

cluster4npu_ui/release_note.md

@@ -0,0 +1,8 @@
+# Release Note for Cluster4NPU
+## v0.0.2 (2025/7/31)
+### Update
+- 
+### Fix
+- Can't inference twice in the same app
+- FPS computation
+- 

cluster4npu_ui/ui/dialogs/deployment.py

@@ -366,7 +366,7 @@ class DeploymentDialog(QDialog):
         
         # Topology tab
         self.topology_tab = self.create_topology_tab()
-        self.tab_widget.addTab(self.topology_tab, "Topology Analysis")
+        self.tab_widget.addTab(self.topology_tab, "Analysis")
         
         # Configuration tab
         self.config_tab = self.create_configuration_tab()