diff --git a/deploy/cpp_infer/src/ocr_det.cpp b/deploy/cpp_infer/src/ocr_det.cpp
index 56fbace8cc6fa27f8172bed248573f15d0c98dac..bf94abce236853410c15434d494058be03a62a81 100644
--- a/deploy/cpp_infer/src/ocr_det.cpp
+++ b/deploy/cpp_infer/src/ocr_det.cpp
@@ -26,6 +26,8 @@ void DBDetector::LoadModel(const std::string &model_dir) {
     config.DisableGpu();
     if (this->use_mkldnn_) {
       config.EnableMKLDNN();
+      // cache 10 different shapes for mkldnn to avoid memory leak
+      config.SetMkldnnCacheCapacity(10);
     }
     config.SetCpuMathLibraryNumThreads(this->cpu_math_library_num_threads_);
   }
diff --git a/deploy/cpp_infer/src/ocr_rec.cpp b/deploy/cpp_infer/src/ocr_rec.cpp
index a3486db46f6eb6ad0df49619744924e6ef70dd01..b997d8291a64f9b6042bce648bcd358e34d55a95 100644
--- a/deploy/cpp_infer/src/ocr_rec.cpp
+++ b/deploy/cpp_infer/src/ocr_rec.cpp
@@ -126,6 +126,8 @@ void CRNNRecognizer::LoadModel(const std::string &model_dir) {
     config.DisableGpu();
     if (this->use_mkldnn_) {
       config.EnableMKLDNN();
+      // cache 10 different shapes for mkldnn to avoid memory leak
+      config.SetMkldnnCacheCapacity(10);
     }
     config.SetCpuMathLibraryNumThreads(this->cpu_math_library_num_threads_);
   }
diff --git a/deploy/cpp_infer/tools/config.txt b/deploy/cpp_infer/tools/config.txt
index 40beea3a2e6f0260a42202d6411ffb10907bf871..6c53f29eeb310677815d106d3e0ae39fb03bc2e2 100644
--- a/deploy/cpp_infer/tools/config.txt
+++ b/deploy/cpp_infer/tools/config.txt
@@ -3,7 +3,7 @@ use_gpu  0
 gpu_id  0
 gpu_mem  4000
 cpu_math_library_num_threads  10
-use_mkldnn 0
+use_mkldnn 1
 use_zero_copy_run 1
 
 # det config
diff --git a/tools/infer/predict_system.py b/tools/infer/predict_system.py
index 647a76b20496335cd059242890f86fffe1e3ac1a..ff5d53e94e8ac110d58f2fda9afeb575cd7f0971 100755
--- a/tools/infer/predict_system.py
+++ b/tools/infer/predict_system.py
@@ -122,7 +122,6 @@ def main(args):
     image_file_list = get_image_file_list(args.image_dir)
     text_sys = TextSystem(args)
     is_visualize = True
-    tackle_img_num = 0
     for image_file in image_file_list:
         img, flag = check_and_read_gif(image_file)
         if not flag:
@@ -131,9 +130,6 @@ def main(args):
             logger.info("error in loading image:{}".format(image_file))
             continue
         starttime = time.time()
-        tackle_img_num += 1
-        if not args.use_gpu and args.enable_mkldnn and tackle_img_num % 30 == 0:
-            text_sys = TextSystem(args)
         dt_boxes, rec_res = text_sys(img)
         elapse = time.time() - starttime
         print("Predict time of %s: %.3fs" % (image_file, elapse))
@@ -153,11 +149,7 @@ def main(args):
             scores = [rec_res[i][1] for i in range(len(rec_res))]
 
             draw_img = draw_ocr(
-                image,
-                boxes,
-                txts,
-                scores,
-                drop_score=drop_score)
+                image, boxes, txts, scores, drop_score=drop_score)
             draw_img_save = "./inference_results/"
             if not os.path.exists(draw_img_save):
                 os.makedirs(draw_img_save)
diff --git a/tools/infer/utility.py b/tools/infer/utility.py
index 8d66f8671815c53d00476033f772cb58069606c8..3e1f07b8a7127e64a994c34d296c945ad1cafd0a 100755
--- a/tools/infer/utility.py
+++ b/tools/infer/utility.py
@@ -101,6 +101,8 @@ def create_predictor(args, mode):
         config.disable_gpu()
         config.set_cpu_math_library_num_threads(6)
         if args.enable_mkldnn:
+            # cache 10 different shapes for mkldnn to avoid memory leak
+            config.set_mkldnn_cache_capacity(10)
             config.enable_mkldnn()
 
     #config.enable_memory_optim()