OpenCV dnn classification result is not match to Caffe result

edit

OpenCV Code:

image = cv2.imread("mysite/27.png")
blob = cv2.dnn.blobFromImage(image, 1, (224, 224), (123, 117, 104), swapRB=False, crop=False)
net = cv2.dnn.readNetFromCaffe("deploy.prototxt", "model.caffemodel")
net.setInput(blob)
preds = net.forward()
idxs = np.argsort(preds[0])[::-1][:5]
print(idxs)

Caffe code:

net = caffe.Classifier(MODEL_FILE, PRETRAINED, mean=np.array([104, 117, 123]), channel_swap=(2,1,0), input_scale=1.0, raw_scale=255, image_dims(224, 224))
input_image = caffe.io.load_image('27.png')
pred = net.predict([input_image])
idxs = np.argsort(pred[0])[::-1][:5]
print(idxs)

Even though I set swapRB=True/False, crop=True/False, all cases return different result. Result on Caffe is my expectation, but opencv isn't. I'm using OpenCV 3.4.2, Caffe 1.0.

Anyone can help. Thanks in advance.