HYTOS

from flask import Flask, Blueprint, jsonify, request, render_template

import cv2

import numpy as np

import sys, os

import json, base64

from PIL import Image

# craft

sys.path.insert(0, os.path.dirname(os.path.realpath(__file__)) + '\\..\\..\\CRAFT_pytorch_master')

# service streamer

sys.path.insert(0, os.path.dirname(os.path.realpath(__file__)) + '\\..\\..\\service_streamer_master')

# deep ocr

# sys.path.insert(0, os.path.dirname(os.path.realpath(__file__)) + '\\deep_text_recognition_benchmark_master')

# symbol

sys.path.insert(0, os.path.dirname(os.path.realpath(__file__)) + '\\..\\..\\symbol_recognition')

recognition_service = Blueprint('recognition', __name__, url_prefix='/recognition')

@recognition_service.route('/', methods=['GET'])

def index():

    return render_template("/recognition/index.html")

def recognition(params=None):

    if params[0][0] == 'Text Area':

        res = text_craft.get_text_box_batch(params[0][1])

        return [res]

    elif params[0] == 'Symbol':

        pass

try:

    # from model import get_prediction, batch_prediction

    import text_craft

    import test_doftech_all_images

    from service_streamer import ThreadedStreamer

    # for error at 3.8

    # import ctypes

    # ctypes.cdll.LoadLibrary('caffe2_nvrtc.dll')

    streamer = ThreadedStreamer(recognition, batch_size=64)

except ImportError as ex:

    ex

    pass

@recognition_service.route('/symbol_box', methods=['POST'])

def symbol_box():

    if request.method == 'POST':

        r = request

        data = json.loads(r.data)

        str_img = base64.b64decode(data['img'])

        nparr = np.fromstring(str_img, np.uint8)

        imgs = []

        imgs.append(Image.fromarray(cv2.imdecode(nparr, cv2.IMREAD_COLOR)))

        data_path = os.path.join(os.path.dirname(os.path.realpath(__file__)) + '\\..\\..\\symbol_training\\Data\\', data['name'])

        boxes = test_doftech_all_images.get_symbol(imgs, data['name'], root_path=data_path, trained_model1=os.path.join(data_path, 'checkpoint', data['name'] + "_only_params.pth"), \

                                                   trained_model2=None, save=False)

        return jsonify({'symbol_box': boxes[0]})

@recognition_service.route('/text_box', methods=['POST'])

def text_box():

    if request.method == 'POST':

        r = request

        nparr = np.fromstring(r.data, np.uint8)

        img = cv2.imdecode(nparr, cv2.IMREAD_COLOR)

        # img = img.reshape(1, -1)

        boxes = text_craft.get_text_box(img, img_path=None, score_path=None, trained_model=os.path.dirname(

            os.path.realpath(__file__)) + '\\CRAFT_pytorch_master\\weights\\craft_mlt_25k.pth')

        return jsonify({'text_box': boxes})

@recognition_service.route('/stream_text_box', methods=['POST'])

def stream_text_box():

    if request.method == 'POST':

        r = request

        str_imgs = json.loads(r.data)

        imgs = []

        for str_img in str_imgs:

            str_img = base64.b64decode(str_img)

            nparr = np.fromstring(str_img, np.uint8)

            img = cv2.imdecode(nparr, cv2.IMREAD_COLOR)

            imgs.append(img)

        boxes_list = []

        '''

        for img in imgs:

            # faster

            #boxes = streamer.predict([[img, None, None, os.path.dirname(os.path.realpath(__file__)) + '\\CRAFT_pytorch_master\\weights\\craft_ic15_20k.pth']])

            # More accurate

            boxes = streamer.predict([[img, None, None, os.path.dirname(os.path.realpath(__file__)) + '\\CRAFT_pytorch_master\\weights\\craft_mlt_25k.pth']])

            boxes_list.append(boxes[0])

        '''

        '''

        infos = []

        for img in imgs:

            infos.append([img, None, None, os.path.dirname(os.path.realpath(__file__)) + '\\CRAFT_pytorch_master\\weights\\craft_mlt_25k.pth'])

        boxes = streamer.predict(infos)

        boxes_list = boxes

        '''

        infos = [['Text Area', [False, os.path.dirname(os.path.realpath(__file__)) + '\\..\\..\\CRAFT_pytorch_master\\weights\\craft_mlt_25k.pth', imgs]]]

        boxes = streamer.predict(infos)

        boxes_list = boxes[0][0]

        return jsonify({'text_box_list': boxes_list})