Dockerfile

Dockerfile

@@ -1,12 +1,10 @@
-FROM  tensorflow/tensorflow:2.5.1
+FROM python:3.7.16
 
-RUN apt-get install make g++ gcc
+RUN pip install gunicorn==20.1.0
-RUN pip3 install gunicorn
+RUN pip install setuptools==46.1.3
-RUN pip3 install setuptools==46.1.3
 
 RUN mkdir -p /app
 WORKDIR /app
-
 COPY requirements.txt /app
 RUN pip install -r requirements.txt -i https://mirrors.aliyun.com/pypi/simple/
 

Dockerfile_arm

@@ -0,0 +1,16 @@
+FROM nvcr.io/nvidia/l4t-tensorflow:r32.6.1-tf2.5-py3
+RUN apt-get install make g++ gcc
+RUN pip3 install gunicorn
+RUN pip3 install setuptools==46.1.3
+
+RUN mkdir -p /app
+WORKDIR /app
+
+COPY requirements.txt /app
+RUN pip3 install -r requirements.txt -i https://mirrors.aliyun.com/pypi/simple/
+
+COPY ./app /app
+EXPOSE 5000
+CMD ["gunicorn", "--bind", ":5000", "server:app"]
+
+

app/serve.py

@@ -1,6 +1,6 @@
 import tensorflow as tf
 import numpy as np
-import keras.models
+
 
 def serve_unet_model():
     TFLITE_MODEL = "../app/UNet_25_Crack.tflite"
@@ -25,9 +25,3 @@ def serve_rcnn_model():
             od_graph_def.ParseFromString(serialized_graph)
             tf.import_graph_def(od_graph_def, name='')
     return detection_graph
-
-
-def serve_bridge_model():
-    mp = "../app/crack_model.h5"
-    model = keras.models.load_model(mp)
-    return model

app/server.py

@@ -8,9 +8,8 @@ from PIL import Image, ImageDraw
 import tensorflow as tf
 import ops as utils_ops
 import visualization_utils as vis_util
-import cv2
 
-from serve import serve_unet_model, serve_bridge_model
+from serve import serve_unet_model
 from serve import serve_rcnn_model
 
 app = Flask(__name__)
@@ -247,95 +246,5 @@ def index():
             return jsonify(data)
     return "Road Damage Detection"
 
-
-@app.route("/predict/bridge", methods=["POST"])
-def bridge():
-    if flask.request.method == "POST":
-        if flask.request.files.get("image"):
-            pred_data_colr = []
-            pred_data_inv = []
-            img_src = cv2.imread(flask.request.files["image"], 0)
-            image_dst = resize_keep_aspect_ratio(img_src, (227, 227))
-            bi_inv, colored_img = process_image(image_dst)
-            pred_data_colr.append(colored_img)
-            pred_data_inv.append(bi_inv)
-            final_pred_colr = np.array(pred_data_colr).reshape((len(pred_data_colr), 227, 227, 1))
-            final_pred_inv = np.array(pred_data_inv).reshape((len(pred_data_inv), 227, 227, 1))
-            is_crack = predict_image_util(final_pred_inv)
-            image_np = load_image_into_numpy_array(img_src)
-            img = Image.fromarray(image_np.astype("uint8"))
-            img = img.resize((128, 128))
-            raw_bytes = io.BytesIO()
-            img.save(raw_bytes, "JPEG")
-            raw_bytes.seek(0)
-            img_byte = raw_bytes.getvalue()
-            img_str = base64.b64encode(img_byte)
-            data = {
-                "result": is_crack,
-                "img": img_str.decode('utf-8')
-            }
-            return jsonify(data)
-        else:
-            data = {
-                "code": 10001,
-                "msg": "Could not find image"
-            }
-            return jsonify(data)
-    return "Bridge Detection"
-
-
-def predict_image_util(final_pred_inv):
-    model = serve_bridge_model()
-    img_test = (final_pred_inv[0].reshape((1, 227, 227, 1)))
-    raw_predicted_label = model.predict(img_test, batch_size=None, verbose=0, steps=None)[0][0]
-
-    predicted_label = 1
-    if raw_predicted_label < 0.8:
-        predicted_label = 0
-
-    predicted_label_str = 'Crack'
-    if predicted_label == 0:
-        predicted_label_str = 'No Crack'
-
-    print('Raw Predicted Label(Numeric): ' + str(raw_predicted_label))
-    print('Predicted Label : ' + predicted_label_str)
-    return predicted_label
-
-
-def process_image(img):
-    ret, bi_inv = cv2.threshold(img, 127, 255, cv2.THRESH_BINARY_INV)
-    return bi_inv, img
-
-
-def resize_keep_aspect_ratio(image_src, dst_size):
-    src_h, src_w = image_src.shape[:2]
-    dst_h, dst_w = dst_size
-
-    # 判断应该按哪个边做等比缩放
-    h = dst_w * (float(src_h) / src_w)  # 按照ｗ做等比缩放
-    w = dst_h * (float(src_w) / src_h)  # 按照h做等比缩放
-
-    h = int(h)
-    w = int(w)
-
-    if h <= dst_h:
-        image_dst = cv2.resize(image_src, (dst_w, int(h)))
-    else:
-        image_dst = cv2.resize(image_src, (int(w), dst_h))
-
-    h_, w_ = image_dst.shape[:2]
-
-    top = int((dst_h - h_) / 2)
-    down = int((dst_h - h_ + 1) / 2)
-    left = int((dst_w - w_) / 2)
-    right = int((dst_w - w_ + 1) / 2)
-
-    value = [0, 0, 0]
-    border_type = cv2.BORDER_CONSTANT
-    image_dst = cv2.copyMakeBorder(image_dst, top, down, left, right, border_type, None, value)
-
-    return image_dst
-
-
 if __name__ == "__main__":
     app.run()

build.sh

@@ -1,3 +1,3 @@
 # /usr/bash
 
-docker build --tag hpds-bridge-detection:1.0.0 .
+docker build --tag hpds-road-detection:1.0.0 .

docker-compose.yml

@@ -1,13 +1,13 @@
 version: "3.6"
 services:
-  hpds-bridge-detection-model:
+  hpds-python-model:
-    container_name: hpds-bridge-detection-model
+    container_name: hpds-road-detection-model
-    image: hpds-bridge-detection:1.0.0
+    image: hpds-road-detection:1.0.0 
     networks:
       - hpds-network
     restart: always
     ports:
-      - "8002:5000"
+      - "8000:5000"
     volumes:
       - /usr/local/cuda/lib64:/usr/local/cuda/lib64
 

install.sh

@@ -1,4 +0,0 @@
-#/bin/bash
-
-pip3 install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple/
-

requirements.txt

@@ -3,5 +3,3 @@ numpy==1.19.5
 Pillow==7.1.2
 six==1.15.0
 tensorflow==2.5.1
-opencv-contrib-python==4.5.3.56
-opencv-python==4.5.3.56