diff --git a/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/README.md b/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/README.md
index bc0a4b3b0e05e6c607a132f9801f3f9ac041e3e9..1c4624fe3e0605d2493817cd4b8cfda21ca66478 100644
--- a/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/README.md
+++ b/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/README.md
@@ -30,8 +30,8 @@ cd Modelzoo-TensorFlow/ACL_TensorFlow/built-in/cv/Facenet_for_ACL
 
 2. 运行预处理脚本
    ```
-   python3 align/align_dataset_mtcnn.py $cur_dir/lfw $dataset
-   python3 preprocess_data.py  Path_of_Data_after_face_alignment  Outpath_of_Data_after_face_alignment  --use_fixed_image_standardization --lfw_batch_size 1
+   python3 align/align_dataset_mtcnn.py $cur_dir/lfw $dataset --image_size 160 --margin 32 --random_order
+   python3 preprocess_data.py  Path_of_Data_after_face_alignment  Outpath_of_Data_after_face_alignment  --use_fixed_image_standardization --lfw_batch_size 1 --use_flipped_images
    
    ```
  
@@ -76,6 +76,6 @@ cd Modelzoo-TensorFlow/ACL_TensorFlow/built-in/cv/Facenet_for_ACL
 
 |       model    |       mode       | ***data***  |    Embeddings Accuracy    |
 | :---------------:| :---------------: | :---------: | :---------: |
-| pb(20180402)| offline Inference | 12000 images |   99.532%     |
-| pb(20180408)| offline Inference | 12000 images |   98.917%     |
+| pb(20180402)| offline Inference | 12000 images |   99.550%     |
+| pb(20180408)| offline Inference | 12000 images |   99.133%     |
 
diff --git a/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/README_EN.md b/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/README_EN.md
index d55ebad700a99ba4d4ad76c70908546637d9ed22..2307ea44d81e6405e8a12c3339314f534a737b1f 100644
--- a/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/README_EN.md
+++ b/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/README_EN.md
@@ -30,8 +30,8 @@ cd Modelzoo-TensorFlow/ACL_TensorFlow/built-in/cv/Facenet_for_ACL
 
 2. Executing the Preprocessing Script
    ```
-   python3 align/align_dataset_mtcnn.py $cur_dir/lfw $dataset
-   python3 preprocess_data.py  Path_of_Data_after_face_alignment  Outpath_of_Data_after_face_alignment  --use_fixed_image_standardization --lfw_batch_size 1
+   python3 align/align_dataset_mtcnn.py $cur_dir/lfw $dataset --image_size 160 --margin 32 --random_order
+   python3 preprocess_data.py  Path_of_Data_after_face_alignment  Outpath_of_Data_after_face_alignment  --use_fixed_image_standardization --lfw_batch_size 1 --use_flipped_images
    
    ```
  
@@ -76,6 +76,6 @@ Our result were obtained by running the applicable inference script. To achieve
 
 |       model    |       mode       | ***data***  |    Embeddings Accuracy    |
 | :---------------:| :---------------: | :---------: | :---------: |
-| pb(20180402)| offline Inference | 12000 images |   99.532%     |
-| pb(20180408)| offline Inference | 12000 images |   98.917%     |
+| pb(20180402)| offline Inference | 12000 images |   99.550%     |
+| pb(20180408)| offline Inference | 12000 images |   99.133%     |
 
diff --git a/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/benchmark_tf.sh b/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/benchmark_tf.sh
index e02318847207963d20a5de7218df4c4003c7fbcd..9d972dac69730ac4e126751941232b0c106d679c 100644
--- a/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/benchmark_tf.sh
+++ b/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/benchmark_tf.sh
@@ -19,9 +19,9 @@ preprocess()
         fi
 	cd $cur_dir/script
         export PYTHONPATH=$cur_dir/script
-	python3 align/align_dataset_mtcnn.py $cur_dir/lfw $dataset
+	python3 align/align_dataset_mtcnn.py $cur_dir/lfw $dataset --image_size 160 --margin 32 --random_order
        
-        python3 preprocess_data.py  $dataset $dataset_bin  --use_fixed_image_standardization --lfw_batch_size 1
+    python3 preprocess_data.py  $dataset $dataset_bin  --use_fixed_image_standardization --lfw_batch_size 1 --use_flipped_images
 }
 infer_test()
 {
@@ -34,7 +34,7 @@ infer_test()
 	#/usr/local/Ascend/atc/bin/atc --framework=3 --model=./model/facenet_tf.pb  --output=./model/facenet --soc_version=Ascend310 --insert_op_conf=./facenet_tensorflow.cfg --input_format=NHWC --input_shape=input:1,160,160,3 
 
 	echo "$cur_dir/Benchmark/out/benchmark --dataDir $dataset_bin/data_image_bin --om $modelPath  --batchSize $batchSize --modelType $modelType --imgType $imgType --deviceId $deviceId --framework $framework --useDvpp $useDvpp > $testcase_dir/performance.log"
-              $cur_dir/Benchmark/out/benchmark --dataDir $dataset_bin/data_image_bin --om $modelPath  --batchSize $batchSize --modelType $modelType --imgType $imgType --deviceId $deviceId --framework $framework --useDvpp $useDvpp > $testcase_dir/performance.log
+    $cur_dir/Benchmark/out/benchmark --dataDir $dataset_bin/data_image_bin --om $modelPath  --batchSize $batchSize --modelType $modelType --imgType $imgType --deviceId $deviceId --framework $framework --useDvpp $useDvpp > $testcase_dir/performance.log
 }
 
 collect_result()
@@ -48,7 +48,7 @@ collect_result()
     echo "InferencePerformance: $AiModel_time ms/batch, $AiModel_throughput images/sec" >> $testcase_dir/performance_results.log
     echo "InferenceTotalTime: $InferenceEngine_total_time ms" >> $testcase_dir/performance_results.log
 	cd $cur_dir/script
-	python3 afterprocess.py  $dataset  $cur_dir/results/${modelType}/ $dataset_bin/data_label_bin/ --lfw_batch_size 1 >  $testcase_dir/file.log
+	python3 afterprocess.py  $dataset  $cur_dir/results/${modelType}/ $dataset_bin/data_label_bin/ --lfw_batch_size 1 --distance_metric 1 --use_flipped_images --subtract_mean >  $testcase_dir/file.log
 	
 	
 	Accu=`cat $testcase_dir/file.log |grep 'Accuracy' | awk '{print $2}'`
diff --git a/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/script/afterprocess.py b/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/script/afterprocess.py
index e3cc4e5509523f84f6c11bfc0d92d9dd1d1ac95b..ed789d6083e81f7e4da915608792033ecb9b305c 100644
--- a/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/script/afterprocess.py
+++ b/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/script/afterprocess.py
@@ -80,23 +80,23 @@ def main(args):
         #feed_dict = {phase_train_placeholder: False, batch_size_placeholder: batch_size}
         #emb, lab = sess.run([embeddings, labels], feed_dict=feed_dict)
         #读取image_bin和label_bin
-        print(paths[i])
-        names = paths[i].split('/')[-1].split('.')[-2]
-        print(names)
-        out_image_name = args.input_dir + "davinci_"+names +'_' + str(i) + "__output0.bin"
-        print(out_image_name)
+        #print(paths[i])
+        #names = paths[i].split('/')[-1].split('.')[-2]
+        #print(names)
+        out_image_name = os.path.join(args.input_dir,"davinci_{}_output0.bin".format(str(i).zfill(6)))
+        #print(out_image_name)
         emb = np.fromfile(out_image_name, dtype="float32").reshape(1, 512)
-        out_label_name = args.label_dir  +names +'_' +str(i) + "_.bin"
+        out_label_name = os.path.join(args.label_dir,"{}.bin".format(str(i).zfill(6)))
         lab = np.fromfile(out_label_name, dtype="int32")
-        print(lab)
+        #print(lab)
         ##########
         lab_array[lab] = lab
         emb_array[lab, :] = emb
         if i % 10 == 9:
             print('.', end='')
             sys.stdout.flush()
-    print('')
-    print(lab_array)
+    #print('')
+    #print(lab_array)
     embeddings = np.zeros((nrof_embeddings, embedding_size * nrof_flips))
     if use_flipped_images:
         # Concatenate embeddings for flipped and non flipped version of the images
diff --git a/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/script/preprocess_data.py b/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/script/preprocess_data.py
index 225e00b4c1bca8985cc4a72abdbbec5eaf271e13..5d839bb5a818290bab2c40deaca94c0a965ec0ad 100644
--- a/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/script/preprocess_data.py
+++ b/ACL_TensorFlow/built-in/cv/Facenet_for_ACL/script/preprocess_data.py
@@ -113,17 +113,15 @@ def evaluate(sess,output_path, enqueue_op,image_paths_placeholder,labels_placeho
     assert nrof_images % batch_size == 0, 'The number of LFW images must be an integer multiple of the LFW batch size'
     nrof_batches = nrof_images // batch_size
     print("#############nrof_batches",nrof_batches)
-    if not os.path.exists(output_path +"data_image_bin/"):
-        os.makedirs(output_path +"data_image_bin/")
-    if not os.path.exists(output_path +"data_label_bin/"):
-        os.makedirs(output_path +"data_label_bin/")
+    if not os.path.exists(os.path.join(output_path ,"data_image_bin/")):
+        os.makedirs(os.path.join(output_path ,"data_image_bin/"))
+    if not os.path.exists(os.path.join(output_path ,"data_label_bin/")):
+        os.makedirs(os.path.join(output_path ,"data_label_bin/"))
     for i in range(nrof_batches):
         ###########save  bin ###############
-        print(i,image_paths[i])
         feed_dict2 = {batch_size_placeholder:batch_size}
-        mid_name = image_paths[i].split('.')[0].split('/')[-1]
-        bin_image2 = output_path +"data_image_bin/" + mid_name + '_'+str(i)+'_'+".bin"
-        bin_label2 = output_path +"data_label_bin/" + mid_name + '_'+str(i)+'_'+".bin"
+        bin_image2 = os.path.join(output_path,"data_image_bin/{}.bin".format(str(i).zfill(6)))
+        bin_label2 = os.path.join(output_path,"data_label_bin/{}.bin".format(str(i).zfill(6)))
         emb, lab  = sess.run([image_batch,labels],feed_dict=feed_dict2)
         emb.astype(np.uint8).tofile(bin_image2)
         lab.tofile(bin_label2)