diff --git a/onnxruntime_cpp_samples/cuda_provider/src/NVIDIAInference.cpp b/onnxruntime_cpp_samples/cuda_provider/src/NVIDIAInference.cpp
index 10176be..cd4cf43 100644
--- a/onnxruntime_cpp_samples/cuda_provider/src/NVIDIAInference.cpp
+++ b/onnxruntime_cpp_samples/cuda_provider/src/NVIDIAInference.cpp
@@ -69,7 +69,7 @@ void NVInference::runAsync(Ort::Value *input, Ort::RunOptions *run_options) {
     /// This event guarantees that we can overwrite the input buffer
     CheckCUDA(cudaEventRecord(m_ev_inference_in, m_compute_stream));
   } else {
-    io_binding.BindOutput(name.get(), *m_input_tensor);
+    io_binding.BindInput(name.get(), *m_input_tensor);
   }