Description: The generated benchmark() method wraps self.predict(), which includes _preprocess_input (OpenCV resize, float conversion). This measures app latency, not model inference latency
Fix: The generated benchmark loop should time only interpreter.invoke().
Description: The generated benchmark() method wraps self.predict(), which includes _preprocess_input (OpenCV resize, float conversion). This measures app latency, not model inference latency
Fix: The generated benchmark loop should time only interpreter.invoke().