app.py

from server.models.image import Image
from server.database.database import Database
from flask import Flask, render_template, request, json, jsonify
import codecs
import tensorflow as tf
import numpy as np
import os



DIR = 'server/cnn_saved_model/saved_model/'
LABELS_ = ["Airplane",
        "Automobile",
        "Bird",
        "Cat",
        "Deer",
        "Dog",
        "Frog",
        "Hourse",
        "Ship",
        "Truck"]

app = Flask(__name__, template_folder="static/")

app.config['MONGO_DBNAME'] = 'gallery_instagram'
app.config['MONGO_URI'] = 'mongodb://pavlo_kuzina:silverok911@ds141406.mlab.com:41406/gallery_instagram'
app.secret_key = "secretverysecetkeyishere"

@app.before_first_request 
def initiliae_database():
    Database.initialize(app)


@app.route("/")
def index(images=[]):
    return render_template("index.html", images=images)


@app.route("/api/get_images")
def get_by_label():
    images = Database.get_all('images')
    list_images = []

    for img in images:
        label_file = img['label']
        fs_file = Database.FS.get(img["fields"])

        base64_data = codecs.encode(fs_file.read(), 'base64')
        image = base64_data.decode('utf-8')
        img_str = "data:image/png;base64," + image
        # jsonImg = json.dumps(img_str)

        list_images.append({"url": img_str, "label": label_file})
    return jsonify(images=list_images)
  

@app.route("/api/get_images/<string:label>")
def get_images_by_label(label):
    batch_images = Image.get_by_label(label)
    # return jsonify(images=batch_images)
    return json.dumps(batch_images)


@app.route("/upload", methods=["POST", "GET"])
def upload_image():
    img_file = request.files['img']

    content_type = img_file.content_type
    filename = img_file.filename

    current_image_id = Image.save_to_mongo(img_file, content_type, filename, label='' )
    
    picture = Database.FS.get(current_image_id)
    saved_image = Image.find_image(current_image_id)

    image = tf.image.decode_jpeg(picture.read(), channels=3)
    image = tf.image.resize_images(image,[ 32, 32])
    
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        tf.get_default_graph().as_graph_def()

        image_value = sess.run([image])
        reshaped_image = np.reshape(image_value[0], [1, 32, 32, 3]).astype(np.float32)

        saver = tf.train.import_meta_graph(os.path.join(DIR,'model-ckpt-4900.meta'))
        saver.restore(sess, os.path.join(DIR,"model-ckpt-4900"))

        x = tf.get_collection('training_data_input')[0]
        y_true = tf.get_collection('training_data_outpuy')[0]
        y_predicted = tf.get_collection('prediction')[0]
        keep_prob =  tf.get_collection('keep_prob')[0]

        pred = sess.run(y_predicted, feed_dict={x: reshaped_image , keep_prob: 1.0})
        sess.close()
        label_idx = np.argmax(pred[0])
        label = LABELS_[label_idx]

    saved_image.set_label(label)

    recognizedImge = Database.FS.get(saved_image.fields)
    base64_data = codecs.encode(recognizedImge.read(), 'base64')
    image = base64_data.decode('utf-8')
    img_str = "data:image/png;base64," + image

    response = {
        "url": img_str,
        "label" : label
    }
    return json.dumps(response)


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