diff --git a/src/operators/kernel/cl/depthwise_conv_kernel.cpp b/src/operators/kernel/cl/depthwise_conv_kernel.cpp
index ad0c9958f817f96139fb93f2a33df0cdec833d35..b84a56ca974fa3aca0f2fe23bb3b9b283eff7a7e 100644
--- a/src/operators/kernel/cl/depthwise_conv_kernel.cpp
+++ b/src/operators/kernel/cl/depthwise_conv_kernel.cpp
@@ -47,7 +47,7 @@ void DepthwiseConvKernel<GPU_CL, float>::Compute(
   int nh = default_work_size[2];
   auto input = param.Input()->GetCLImage();
   auto filter = param.Filter()->GetCLImage();
-  auto output = param.Output();
+  auto output = param.Output()->GetCLImage();
   int stride = param.Strides()[0];
   int offset = param.Offset();
   int input_c = param.Input()->CBlock();