Building a Keras + deep learning REST API（三部曲之一）

时间：2018-06-20 23:58:22 阅读：381 评论：0 收藏：0 [点我收藏+]

标签：ssi art rop post www. 分享 __name__ convert curl

一、基本环境

$ pip install flask gevent requests pillow

其中 flask不需要解释

gevent 是用于自动切换进程的；

pillow 是用来进行python下的图像处理的；

requests 是用来进行python下request处理的。

二、核心代码解释

# import the necessary packages
from keras.applications import ResNet50
from keras.preprocessing.image import img_to_array
from keras.applications import imagenet_utils
from PIL import Image
import numpy as np
import flask
import io

引入所需的头文件。其中注意keras的几个类库是很有通用性的；

# initialize our Flask application and the Keras model
app = flask.Flask(__name__)
model = None

类库的初始化

def load_model():
    # load the pre-trained Keras model (here we are using a model
    # pre-trained on ImageNet and provided by Keras, but you can
    # substitute in your own networks just as easily)
    global model
    model = ResNet50(weights="imagenet")

引入model模型，如果想引入自己的模型(CBIR)的话，就在这里引入。

def prepare_image(image, target):
    # if the image mode is not RGB, convert it
    if image.mode != "RGB":
        image = image.convert("RGB")

    # resize the input image and preprocess it
    image = image.resize(target)
    image = img_to_array(image)
    image = np.expand_dims(image, axis=0)
    image = imagenet_utils.preprocess_input(image)

    # return the processed image
    return image

image的预处理，这里使用的是keras+PIL，和opencv之间的比较，需要有时间来做。

@app.route("/predict", methods=["POST"])
def predict():
    # initialize the data dictionary that will be returned from the
    # view
    data = {"success": False}

    # ensure an image was properly uploaded to our endpoint
    if flask.request.method == "POST":
        if flask.request.files.get("image"):
            # read the image in PIL format
            image = flask.request.files["image"].read()
            image = Image.open(io.BytesIO(image))

            # preprocess the image and prepare it for classification
            image = prepare_image(image, target=(224, 224))

            # classify the input image and then initialize the list
            # of predictions to return to the client
            preds = model.predict(image)
            results = imagenet_utils.decode_predictions(preds)
            data["predictions"] = []

            # loop over the results and add them to the list of
            # returned predictions
            for (imagenetID, label, prob) in results[0]:
                r = {"label": label, "probability": float(prob)}
                data["predictions"].append(r)

            # indicate that the request was a success
            data["success"] = True

    # return the data dictionary as a JSON response
    return flask.jsonify(data)

虽然是核心部分，但是其实非常容易被复用。就是读取数据，然后进行处理的过程。

# if this is the main thread of execution first load the model and
# then start the server
if __name__ == "__main__":
    print(("* Loading Keras model and Flask starting server..."
        "please wait until server has fully started"))
    load_model()
    app.run()