diff --git a/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_fp16.cc b/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_fp16.cc
index e829552771..ad15ac2374 100644
--- a/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_fp16.cc
+++ b/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_fp16.cc
@@ -70,10 +70,6 @@ int DeConvolutionFp16CPUKernel::InitWeightBias() {
       MS_LOG(ERROR) << "DeConv fp16 only support fp16 weight";
       return RET_ERROR;
     }
-    if (bias_size != in_tensors_.at(2)->Size()) {
-      MS_LOG(ERROR) << "input bias size not match : " << bias_size << " vs " << in_tensors_.at(2)->Size();
-      return RET_ERROR;
-    }
     memcpy(bias_data_, in_tensors_.at(2)->data_c(), bias_size);
   }
 
diff --git a/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_fp16.h b/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_fp16.h
index 3256bb75ad..da0330a295 100644
--- a/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_fp16.h
+++ b/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_fp16.h
@@ -50,12 +50,12 @@ class DeConvolutionFp16CPUKernel : public ConvolutionBaseCPUKernel {
   int output_plane_;
   int thread_count_;
   int thread_stride_;
-  float16_t *pack_input_;
-  float16_t *pack_weight_;
-  float16_t *pack_output_;
-  float16_t *tmp_buffer_;
-  float16_t *batch_input_;
-  float16_t *batch_output_;
+  float16_t *pack_input_ = nullptr;
+  float16_t *pack_weight_ = nullptr;
+  float16_t *pack_output_ = nullptr;
+  float16_t *tmp_buffer_ = nullptr;
+  float16_t *batch_input_ = nullptr;
+  float16_t *batch_output_ = nullptr;
 };
 }  // namespace mindspore::kernel
 #endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP16_DECONVOLUTION_H_