ValueError：张量“ A”必须与张量“ B”来自同一图

Question

我正在使用keras的预训练模型，调用ResNet50（weights ='imagenet'）时出现错误。我在Flask服务器中有以下代码：

def getVGG16Prediction(img_path):

    model = VGG16(weights='imagenet', include_top=True)
    img = image.load_img(img_path, target_size=(224, 224))
    x = image.img_to_array(img)
    x = np.expand_dims(x, axis=0)
    x = preprocess_input(x)

    pred = model.predict(x)
    return sort(decode_predictions(pred, top=3)[0])


def getResNet50Prediction(img_path):

    model = ResNet50(weights='imagenet') #ERROR HERE
    img = image.load_img(img_path, target_size=(224, 224))
    x = image.img_to_array(img)
    x = np.expand_dims(x, axis=0)
    x = preprocess_input(x)

    preds = model.predict(x)
    return decode_predictions(preds, top=3)[0]

在main中调用时，它工作正常

if __name__ == "__main__":
    STATIC_PATH = os.getcwd()+"/static"
    print(getVGG16Prediction(STATIC_PATH+"/18.jpg"))
    print(getResNet50Prediction(STATIC_PATH+"/18.jpg"))

但是，当我从烧瓶POST函数调用它时，ValueError上升：

@app.route("/uploadMultipleImages", methods=["POST"])
def uploadMultipleImages():
    uploaded_files = request.files.getlist("file[]")
    weight = request.form.get("weight")

    for file in uploaded_files:
        path = os.path.join(STATIC_PATH, file.filename)
        file.save(os.path.join(STATIC_PATH, file.filename))
        result = getResNet50Prediction(path)

完整的错误如下：

ValueError：Tensor（“ cond / pred_id：0”，dtype = bool）必须与Tensor（“ batchnorm / add_1：0”，shape =（?, 112，112，64），dtype = float32）来自同一张图

任何意见或建议均受到高度赞赏。谢谢。

Answer 1

您将需要打开不同的会话，并指定每个会话使用哪个图，否则Keras将默认替换每个图。

from tensorflow import Graph, Session, load_model
from Keras import backend as K

加载图形：

graph1 = Graph()
    with graph1.as_default():
        session1 = Session()
        with session1.as_default():
            model = load_model(foo.h5)

graph2 = Graph()
    with graph2.as_default():
        session2 = Session()
        with session2.as_default():
            model2 = load_model(foo2.h5)

预测/使用图形：

K.set_session(session1)
    with graph1.as_default():
        result = model.predict(data)