0

I made an ML model that classifies images, and I want to implement it to my application. But the problem is that the file is too heavy and requires an unreasonable amount of disk space for an android application. So I was wondering if I could host the Model function in a server, which would return the prediction back to the client, when I can call from my python script. I couldn't find any specific tutorials on hosting functions, with would return output back the the client. So How do I go about doing this?

Here is the function I wish to host-

from keras.models import load_model
from PIL import Image, ImageOps
import numpy as np
labels= ["Banana", "Fan", "Clock","Coin","Leaf","Paper_airplane","Pen","Phone","Spoon","Tomato"]
model = load_model('keras_model.h5')
data = np.ndarray(shape=(1, 224, 224, 3), dtype=np.float32)

#Here I wish to pass the image, from my python program, and recieve the prediction
def RunPrediction(img):
    image = img
    size = (224, 224)
    image = ImageOps.fit(image, size, Image.ANTIALIAS)

    image_array = np.asarray(image)
    normalized_image_array = (image_array.astype(np.float32) / 127.0) - 1
    data[0] = normalized_image_array

    #And I want to recieve this output in my code
    prediction = model.predict(data)

    return prediction
davidism
  • 121,510
  • 29
  • 395
  • 339
CatCoder
  • 39
  • 5

1 Answers1

0

it is not a good idea to predict on a request! it should answer the user under 1 sec. so better to create model and save it on the disk. after receiving a req, you can only evaluate and pass the tag

omid mesgarha
  • 23
  • 6
  • It's okay to make predictions using an existing model and serve using the endpoint. There is no such constraint to predict within 1 sec. – Rahul Kumar Dec 05 '21 at 09:46