diff --git a/src/kernels/cuda/batch_norm.cc b/src/kernels/cuda/batch_norm.cc
index 7ca75ba6..b150aaa5 100644
--- a/src/kernels/cuda/batch_norm.cc
+++ b/src/kernels/cuda/batch_norm.cc
@@ -28,9 +28,11 @@ class BatchNormCudnn : public CudaKernelWithoutConfig {
         for (size_t i = 0; i < dims.size(); ++i) {
             dimArray[i] = dims[i];
             strideArray[i] = op->getInputs(0)->getStride()[i];
-            dimPArray[i] = op->getInputs(1)->getDims()[i];
-            stridePArray[i] = op->getInputs(1)->getStride()[i];
+            dimPArray[i] = 1;
+            stridePArray[i] = 1; 
         }
+        dimPArray[1] = op->getInputs(0)->getDims()[1];
+        stridePArray[1] = op->getInputs(0)->getStride()[1];
         // get inputs
         cudnnTensorDescriptor_t inDesc;
         checkCudnnError(cudnnCreateTensorDescriptor(&inDesc));