api/views.py

from django.shortcuts import render
from django.http import JsonResponse
import tensorflow as tf
import tensorflow_hub as hub
from tensorflow import keras
from PIL import Image,ImageOps
import numpy as np
import cv2 as cv
from tensorflow.keras import preprocessing
from tensorflow.keras.models import load_model
from tensorflow.keras.activations import softmax
from sklearn.preprocessing import OneHotEncoder
import os
import h5py
from django.views.decorators.csrf import csrf_exempt

model = tf.keras.models.load_model('api/mlModel/model.h5')
shape = ((50,50,3))
model = tf.keras.Sequential([hub.KerasLayer(model,input_shape=shape)])

modelV2 = tf.keras.models.load_model('api/mlModel/model2.0.h5')

@csrf_exempt 
def predict(request):
    if request.method == 'POST':
        
        #image = Image.open("api/mlModel/0.jpg")

        # Get the image from the request
        print(request.FILES['image'])
        image = Image.open(request.FILES['image'])

        # Preprocess the image
        
        test_image = image.resize((50,50))
        test_image = preprocessing.image.img_to_array(test_image)
        test_image = test_image / 255
        test_image = np.expand_dims(test_image, axis =0)
        class_names = ['1','2','3' ,'4', '5', '6' ,'7' ,'8' ,'9' ,'A' ,'B' ,'C' ,'D' ,'E' ,'F', 'G', 'H', 'I', 'J', 'K' ,'L' ,'M', 'N', 'O' ,'P' ,'Q', 'R' ,'S', 'T', 'U' ,'V' ,'W' ,'X' ,'Y' ,'Z']

        # Make a prediction
        predictions = model.predict(test_image)
        scores = tf.nn.softmax(predictions[0])
        scores = scores.numpy()
        image_class = class_names[np.argmax(scores)]
        print(image_class)

        return JsonResponse({'prediction': image_class})
    
    else:
        return render(request,'predict.html')

    
# Create your views here.

def form_view(request):
    return render(request,'predict.html')

@csrf_exempt 
def predictV2(request):
    if request.method == 'POST':
        
        #image = Image.open("api/mlModel/0.jpg")

        # Get the image from the request
        print(request.FILES['image'])
        #image1 = Image.open(request.FILES['image'])
        #img = cv.imread(Image.open(request.FILES['image']))
        img = cv.imdecode(np.fromstring(request.FILES['image'].read(), np.uint8), cv.IMREAD_UNCHANGED)
        # Preprocess the image
        
        resized_img = cv.resize(img, (250, 250), interpolation=cv.INTER_CUBIC)
        resized_img.shape
        #plt.imshow(resized_img)
        img = resized_img
        pred = modelV2.predict(x = np.array(img).reshape(-1,250,250,3)).flatten()
        
        enc = OneHotEncoder()
        enc.fit([['6'],
                ['K'],
                ['L'],
                ['R'],
                ['V'],
                ['3'],
                ['F'],
                ['M'],
                ['J'],
                ['0'],
                ['9'],
                ['U'],
                ['8'],
                ['P'],
                ['W'],
                ['Q'],
                ['N'],
                ['E'],
                ['Y'],
                ['H'],
                ['1'],
                ['X'],
                ['C'],
                ['G'],
                ['5'],
                ['O'],
                ['S'],
                ['B'],
                ['2'],
                ['7'],
                ['D'],
                ['T'],
                ['4'],
                ['I'],
                ['A'],
                ['Z']])
        out = enc.inverse_transform(pred.reshape(1,-1))
        print(out[0][0])


        return JsonResponse({'prediction': out[0][0]})
    
    else:
        return render(request,'predict.html')