app.py

from flask             import Flask, render_template, jsonify, request, Response
from PIL               import Image, ImageOps 
from io                import BytesIO
from sklearn.externals import joblib
import pickle
import json
import pickle
import numpy as np
import requests

# These are the possible categories (classes) which can be detected
namemap = [
    'axes',
    'boots',
    'carabiners',
    'crampons',
    'gloves',
    'hardshell_jackets',
    'harnesses',
    'helmets',
    'insulated_jackets',
    'pulleys',
    'rope',
    'tents'
]
app = Flask(__name__)


def resize(image):
    """ Resize any image to 128 x 128, which is what the model has been trained on """
    
    base = 128
    width, height = image.size

    if width > height:
        wpercent = base/float(width)
        hsize    = int((float(height) * float(wpercent)))
        image    = image.resize((base,hsize), Image.ANTIALIAS)
    else:
        hpercent = base/float(height)
        wsize    = int((float(width) * float(hpercent)))
        image    = image.resize((wsize, base), Image.ANTIALIAS)

    newImage = Image.new('RGB',
                     (base, base),     # A4 at 72dpi
                     (255, 255, 255))  # White

    position = (int( (base/2 - image.width/2) ), 0)
    newImage.paste(image, position)

    return newImage


def normalize(arr):
    """ This means that the largest value for each attribute is 1 and the smallest value is 0.
        Normalization is a good technique to use when you do not know the distribution of your data 
        or when you know the distribution is not Gaussian (a bell curve)."""
    
    arr = arr.astype('float')
    
    # Do not touch the alpha channel
    for i in range(3):
        minval = arr[...,i].min()
        maxval = arr[...,i].max()
        if minval != maxval:
            arr[...,i] -= minval
            arr[...,i] *= (255.0/(maxval-minval))

    return arr


def processImage(image):
    """ Resize and normalize the image """

    image   = resize(image)
    arr     = np.array(image)
    new_img = Image.fromarray(normalize(arr).astype('uint8'),'RGB')
    
    return new_img


@app.route('/classify', methods=['POST'])
def classify():
    """ Make a POST to this endpint and pass in an URL to an image in the body of the request.
        Swap out the model.pkl with another trained model to classify new objects. """
    
    try:
        body    = request.get_json()
        print(body)
        img_url = body["url"]

        # Get the image and show it
        response = requests.get(img_url)
        img      = Image.open(BytesIO(response.content))
        prcedImg = processImage(img)

        # Convert a 2D image into a flat array
        imgFeatures = np.array(prcedImg).ravel().reshape(1,-1)
        print(imgFeatures)

        model   = joblib.load('pickle_model.pkl')
        predict = model.predict(imgFeatures)
        print('The image is a ', namemap[int(predict[0])]),
        
        print('image: ', namemap[int(predict[0])], predict)

        # Convert the integer returned from the model into the name of the class from our namemap above.
        # EX: 0 = axes, 1 = boots, 2 = carabiners
        response = json.dumps({"classification": namemap[int(predict[0])]})
        return(response)
        
    except Exception as e:
        print(e)
        raise 

if __name__ == '__main__':
    app.run(debug=True,host='0.0.0.0')