flask_serve.py

# from crypt import methods
from flask import Flask, jsonify, request
import os
import requests
import numpy as np
from numpy.linalg import norm
import PIL
import time
import math
import tensorflow as tf
from tensorflow.keras.preprocessing import image
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.applications.efficientnet import EfficientNetB5, preprocess_input
from sklearn.neighbors import NearestNeighbors
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
from sklearn.decomposition import PCA
from flask_cors import CORS
import urllib.request
import time


os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3"
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"

app = Flask(__name__)
CORS(app)

@app.route('/pd/', methods=["GET"])
def predeict_image():
    start = time.time()
    img_url = request.args.get("img")
    date = request.args.get("happenDate")
    if img_url != None and date != None:
        path = '/home/aimungtage/images/announcement'
        file_list = os.listdir(path)
        target_list = []
        data_number = 0
        for i in file_list:
            if date <= i:
                target_list.append(i)

        for i in target_list:
            data_path = '/home/aimungtage/images/announcement/' + str(i)
            data_number += len(os.listdir(data_path))

        if data_number == 0:
            k_number = 0
            components = 0
            exit()
        elif data_number <= 10:
            k_number = data_number
            components = 4

        elif data_number <= 50:
            k_number = data_number
            components = 10
        else:
            k_number = 100
            components = 18

        image_path = "/home/aimungtage/images/announcement"

        batch_size = 8
        img_generator = ImageDataGenerator(preprocessing_function = preprocess_input)
        data_generator = img_generator.flow_from_directory(image_path, target_size = (img_size,img_size), batch_size = batch_size, class_mode = None, shuffle=False, classes=target_list)

        num_images = data_generator.samples
        num_epochs = int(math.ceil(num_images / batch_size))

        feature_list = model_resnet.predict(data_generator, num_epochs)

        filenames = [image_path + '/' + s for s in data_generator.filenames]
        pca = PCA(n_components = components)
        pca.fit(feature_list)
        compressed_features = pca.transform(feature_list)

        neighbors = NearestNeighbors(n_neighbors=k_number, algorithm='ball_tree', metric='euclidean',radius = 1.6)
        neighbors.fit(compressed_features)

        search_image_url = img_url.split("?")[0]
        responds = requests.get(search_image_url)
        search_image_name = search_image_url.split("/")[-1]
        open(f"{search_image_name}", "wb").write(responds.content)
        print(search_image_name)

        search_file = search_image_name
        input_shape = (img_size, img_size, 3)
        img = image.load_img(search_file, target_size=(input_shape[0], input_shape[1]))
        img_array = image.img_to_array(img)
        expanded_img_array = np.expand_dims(img_array, axis=0)
        preprocessed_img = preprocess_input(expanded_img_array)
        test_img_features = model_resnet.predict(preprocessed_img, batch_size=1)
        image_test = pca.transform(test_img_features)
        distance, indices = neighbors.kneighbors(image_test)
        outputs = {}

        def similar_images(indices):
            number = 1
            for index in indices:
                if number <=len(indices):
                    outputs[number-1]=os.path.splitext(os.path.basename(filenames[index]))[0]
                    number +=1

        similar_images(indices[0])
        end = time.time()
        print(f" 소요 시간 : {end - start}\n")

    return jsonify(outputs)

if __name__ == "__main__":
  img_size = 256 #input size
  model_resnet = EfficientNetB5(weights='imagenet', include_top=False,input_shape=(img_size, img_size, 3),pooling='max')
  app.run(host="0.0.0.0", port=5000, debug=True)