diff --git a/dnmetis/README.md b/dnmetis/README.md
new file mode 100644
index 0000000..43851e7
--- /dev/null
+++ b/dnmetis/README.md
@@ -0,0 +1,35 @@
+## 1.Install dnmetis_backend
+
+As README.md:
+backend_C++/dnmetis_backend/README.md
+
+## 2.Download dataset and model(.om)
+
+1.download  Imagenet-val dataset URL: http://www.image-net.org/download-imageurls
+
+2.download  efficientnet-b8 model(.om) URL: obs://hwwheel23/efficientnet-b8.om 
+
+3.process the original Imagenet-val dataset as list:
+
+![输入图片说明](https://images.gitee.com/uploads/images/2020/0918/234302_a572d632_5418572.jpeg "无标题.jpg")
+
+
+
+## 3.Start execute the inference:
+
+sh run_efficientnet-b8.sh
+
+or 
+
+python3.7 main.py --model=/data/hwwheel123/model/efficientnet-b8.om --image_size='672,672,3' --inputs='images:0' --outputs='Softmax:0' --precision=fp16
+
+
+## 4.ATC offline model generate (optional):
+
+1.download  efficientnet-b8 model(.pb) URL: obs://hwwheel23/efficientnet-b8.pb 
+
+2.atc --model=$MODEL_DIR/efficientnet-b8.pb --framework=3 --input_shape='images:1,672,672,3' --output=$MODEL_DIR/efficientnet-b8 --mode=0 --out_nodes='Softmax:0' --soc_version=Ascend310  --input_fp16_nodes=images --output_type=FP16
+
+## 5.Imagenet2012-val Top1 Accuracy:
+
+![输入图片说明](https://images.gitee.com/uploads/images/2020/0919/010210_5cf496fc_5418572.png "屏幕截图.png")
diff --git a/dnmetis/main.py b/dnmetis/main.py
new file mode 100644
index 0000000..2c5880b
--- /dev/null
+++ b/dnmetis/main.py
@@ -0,0 +1,150 @@
+import argparse
+import array
+import collections
+import json
+import os
+import sys
+import threading
+import time
+from queue import Queue
+#import env
+import cv2
+import numpy as np
+import re
+import pdb
+
+# import converter.converter as converter
+#from backend.backend_acl import AclBackend
+
+
+
+last_timing = []
+last_device_timing = []
+
+
+def get_args():
+    """Parse commandline."""
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--dataset_path", default="./datasets/imagenet_10", help="path to the dataset")
+    parser.add_argument("--backend", default="acl", help="runtime to use")
+    parser.add_argument("--model", required=True, help="model file path")
+    parser.add_argument("--image_size",default='224,224,3',help="model inputs imagesize")
+    parser.add_argument("--inputs", default="input:0", help="model inputs nodes eg: data1:0 ")
+    parser.add_argument("--outputs",default="softmax:0", help="model outputs nodes list eg:fc1:0,fc2:0,fc3:0 ")
+
+    # below will override DNMetis rules compliant settings - don't use for official submission
+    parser.add_argument("--count", default=1000, type=int, help="dataset items to infer")
+    parser.add_argument("--precision", default="fp32", choices=["fp32", "fp16", "int8", "uint8"],
+                         help="precision mode, one of " + str(["fp32", "fp16", "int8", "uint8"]))
+    parser.add_argument("--feed", default=[], help="feed")
+    parser.add_argument("--image_list", default=[], help="image_list")
+    parser.add_argument("--label_list", default=[], help="label_list")
+    parser.add_argument("--cfg_path",default="./backend_cfg/built-in_config.txt")
+    args = parser.parse_args()
+
+    # don't use defaults in argparser. Instead we default to a dict, override that with a profile
+    # and take this as default unless command line give
+
+
+    #if args.image_size is None:
+    #    args.image_size = SUPPORTED_DATASETS[args.dataset][3]['image_size']
+
+    if args.inputs:
+        args.inputs = args.inputs.split(",")
+    if args.outputs:
+        args.outputs = args.outputs.split(",")
+    if args.image_size:
+        args.image_size = list(map(int, args.image_size.split(",")))
+
+    return args
+
+def get_backend(backend):
+    if backend == "acl":
+        from backend.backend_acl import AclBackend
+        backend = AclBackend()
+    return backend
+
+def resize_with_aspectratio(img, out_height, out_width, scale=87.5, inter_pol=cv2.INTER_LINEAR):
+    height, width = img.shape[:2]
+    new_height = int(100. * out_height / scale)
+    new_width = int(100. * out_width / scale)
+    if height > width:
+        w = new_width
+        h = int(new_height * height / width)
+    else:
+        h = new_height
+        w = int(new_width * width / height)
+    img = cv2.resize(img, (w, h), interpolation=inter_pol)
+    return img
+
+def center_crop(img, out_height, out_width):
+    height, width = img.shape[:2]
+    left = int((width - out_width) / 2)
+    right = int((width + out_width) / 2)
+    top = int((height - out_height) / 2)
+    bottom = int((height + out_height) / 2)
+    img = img[top:bottom, left:right]
+    return img
+def pre_process_noisy(img, dims=None, precision="fp32"):
+    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
+    output_height, output_width, _ = dims
+    cv2_interpol = cv2.INTER_CUBIC
+    img = resize_with_aspectratio(img, output_height, output_width, inter_pol=cv2_interpol)
+    img = center_crop(img, output_height, output_width)
+    MEAN_RGB = [0.485 * 255, 0.456 * 255, 0.406 * 255]
+    STDDEV_RGB = [0.229 * 255, 0.224 * 255, 0.225 * 255]
+
+    if precision=="fp32":
+        img = np.asarray(img, dtype='float32')
+    if precision=="fp16":
+        img = np.asarray(img, dtype='float16')
+
+    means = np.array([0.485 * 255, 0.456 * 255, 0.406 * 255], dtype=np.float32)
+    img -= means
+    stddev = np.array([0.229 * 255, 0.224 * 255, 0.225 * 255], dtype=np.float32)
+    img /= stddev
+    return img
+
+
+
+def preprocess_dataset(args,offset=0):
+    with open(args.dataset_path + '/val_map.txt', 'r') as f:
+        for s in f:
+            image_name, label = re.split(r"\s+", s.strip())
+            src = os.path.join(args.dataset_path, image_name)
+            img_org = cv2.imread(src)
+            processed_img = pre_process_noisy(img_org, dims=args.image_size, precision = args.precision)
+            args.feed.append(processed_img)
+            args.image_list.append(image_name)
+            args.label_list.append(int(label)+offset)
+
+
+def main():
+    good = 0
+    total = 0
+    #args
+    args = get_args()
+
+    # find backend
+    backend = get_backend(args.backend)
+
+    # load model to backend
+    model = backend.load(args)
+    #
+    # preprocess_dataset
+    preprocess_dataset(args,offset=0)
+
+    #start inference:
+    for i in range(len(args.feed)):
+        predictions = backend.predict(args.feed[i])
+        #print(args.feed[i].shape)
+        print('img_orig:',args.image_list[i],'label:',args.label_list[i],'predictions:',np.argmax(predictions),'\n')
+        if args.label_list[i] == np.argmax(predictions):
+            good +=1
+        total +=1
+    print('Predict total jpeg:',len(args.image_list),' Accuracy: ',good / total)
+
+    backend.unload()
+
+if __name__ == "__main__":
+    main()
diff --git a/dnmetis/run_inference.sh b/dnmetis/run_inference.sh
new file mode 100644
index 0000000..2cbd1f7
--- /dev/null
+++ b/dnmetis/run_inference.sh
@@ -0,0 +1,4 @@
+python3.7 main.py --model=./model/efficientnet-b8.om --image_size='672,672,3' --precision=fp16
+
+
+
diff --git a/dnmetis/version.py b/dnmetis/version.py
new file mode 100644
index 0000000..2432e3c
--- /dev/null
+++ b/dnmetis/version.py
@@ -0,0 +1,2 @@
+
+version = '1.0'