app.py

# from __future__ import division, print_function
# coding=utf-8
# import sys
import os
# import glob
# import re
import numpy as np
# import datetime
# Keras
# from tensorflow.keras.models import load_model
#from tensorflow.keras.preprocessing import image

# Flask utils
# from flask import Flask, redirect, url_for, request, render_template
# from werkzeug.utils import secure_filename
# from gevent.pywsgi import WSGIServer





#import everytnimg
# from skimage.io import imread, imshow
# from skimage.filters import gaussian, threshold_otsu
# from skimage.feature import canny
# from skimage.transform import probabilistic_hough_line, rotate
# from process_image import process_image

# import glob
# import math

import cv2 
# import numpy as np
# from PIL import Image
# from matplotlib import pyplot as plt
# from matplotlib.patches import Rectangle
#%matplotlib inline 



# from collections import OrderedDict
# from PIL import Image

# import pandas as pd
# import seaborn as sns

# import math


#import all from Hough transfrom cell
# from skimage.transform import hough_line, hough_line_peaks
# from skimage.transform import rotate
# from skimage.feature import canny
# from skimage.io import imread
# from skimage.color import rgb2gray
# import matplotlib.pyplot as plt
# from scipy.stats import mode as md
# from myhough import deskew, deskew2

# from segment_words import sortit,words,createk,hpf,bps,wps,baw

# from myverify import verify
#from detect_frame import detect_frame
# import pathlib
from PIL import ImageFont, ImageDraw, Image
font = ImageFont.truetype("kalpurush.ttf", 60) #https://img.shields.io/badge/IEEE-10499463-0072bc.svg

citation_text = """<div>
                    <br><p>This is a demo space for the paper: <i>Improving Character Recognition in Bangla Handwritten Words: A Two-Stage Single Shot Detector Approach</i>.</p>
                    <a href='https://ieeexplore.ieee.org/document/10499463' style='text-decoration: none;'>
                        <img src='https://img.shields.io/static/v1?label=%E2%80%8E&logo=ieee&logoSize=auto&message=10499463&color=0072bc&labelColor=323634' alt='IEEE Badge' style='max-width: 100%; height: auto; margin-bottom: 10px;'>
                    </a>
                    <p>Direct Link to the Paper: 
                    <a href='https://ieeexplore.ieee.org/document/10499463' style='text-decoration: none; color: blue;'>https://ieeexplore.ieee.org/document/10499463</a></p>
                    <p>ResearchGate Link to the Paper: 
                    <a href='https://www.researchgate.net/publication/380009779_Improving_Character_Recognition_in_Bangla_Handwritten_Words_A_Two-Stage_Single_Shot_Detector_Approach' style='text-decoration: none; color: blue;'>https://bit.ly/ResearchGatePaperLink</a></p>
                    <p>Please cite this paper with BibTeX as follows:</p>
                    </div>
                    <div>
    <pre><code style="overflow-x: auto;">@inproceedings{pal2024improving,
  title={Improving Character Recognition in Bangla Handwritten Words: A Two-Stage Single Shot Detector Approach},
  author={Pal, Avi and Hasan, Md Sajid and Ahsan, Sk Md Masudul},
  booktitle={2024 International Conference on Advances in Computing, Communication, Electrical, and Smart Systems (iCACCESS)},
  pages={1--6},
  year={2024},
  organization={IEEE}
}
</code></pre>
    </div>"""

#import more
import tensorflow as tf
from object_detection.utils import config_util
# from object_detection.protos import pipeline_pb2
# from google.protobuf import text_format

# import os
from object_detection.utils import label_map_util
# from object_detection.utils import visualization_utils as viz_utils
from object_detection.builders import model_builder

# Load pipeline config and build a detection model
WORKSPACE_PATH = 'Tensorflow/workspace'
# SCRIPTS_PATH = 'Tensorflow/scripts'
#APIMODEL_PATH = 'Tensorflow/models'
ANNOTATION_PATH = WORKSPACE_PATH+'/annotations'
# IMAGE_PATH = WORKSPACE_PATH+'/images'
MODEL_PATH = WORKSPACE_PATH+'/models'
PRETRAINED_MODEL_PATH = WORKSPACE_PATH+'/pre-trained-models'
CONFIG_PATH = MODEL_PATH+'/my_ssd_mobnet/pipeline.config'
CHECKPOINT_PATH = MODEL_PATH+'/my_ssd_mobnet/'
# INPUT_IMAGE_PATH = 'Tensorflow/myimages'
# MODEL_PATH = 'E:/RealTimeObjectDetection/model.best.hdf5'

configs = config_util.get_configs_from_pipeline_file(CONFIG_PATH)
detection_model = model_builder.build(model_config=configs['model'], is_training=False)

# Restore checkpoint
ckpt = tf.compat.v2.train.Checkpoint(model=detection_model)
ckpt.restore(os.path.join(CHECKPOINT_PATH, 'ckpt-51')).expect_partial()

@tf.function
def detect_fn(image):
    image, shapes = detection_model.preprocess(image)
    prediction_dict = detection_model.predict(image, shapes)
    detections = detection_model.postprocess(prediction_dict, shapes)
    return detections

def detect_frame(frame,isRealTime = False):
    image_np = np.array(frame)
    cpimg = frame.copy()
    input_tensor = tf.convert_to_tensor(np.expand_dims(image_np, 0), dtype=tf.float32)
    detections = detect_fn(input_tensor)
    print(len(detections))
    num_detections = int(detections.pop('num_detections'))
    #print("hello")
    #print(num_detections)
    
    #print(len(detections['detection_scores']))
    detections = {key: value[0, :num_detections].numpy()
                  for key, value in detections.items()}
    detections['num_detections'] = num_detections
    row,col,dummy = image_np.shape
    # detection_classes should be ints.
    detections['detection_classes'] = detections['detection_classes'].astype(np.int64)
    #print(detections['detection_classes'])
    mark = [0]*15
    myletters = []
    for i in range(0,15):
        curi=detections['detection_classes'][i]
        classi=classes[curi]
        print(classes[curi],end='-')
        cur=detections['detection_scores'][i]
        if(cur<0.2):
            continue
        print(cur,end=' ')
        print(detections['detection_boxes'][i], end=' ')
        x0=(detections['detection_boxes'][i][0])
        y0=(detections['detection_boxes'][i][1])
        x1=(detections['detection_boxes'][i][2])
        y1=(detections['detection_boxes'][i][3])
        curarea=(x1-x0)*(y1-y0)
        ok=1
        for j in range(0,i):
            #print(mark[j])
            if mark[j]==0:
                continue
            curj=detections['detection_classes'][j]
            classj=classes[curj]
            
            if classi=='ি' or classj=='ি':
                if classi!=classj:
                    continue
            if classi=='ী' or classj=='ী':
                if classi!=classj:
                    continue
            
            x2=(detections['detection_boxes'][j][0])
            y2=(detections['detection_boxes'][j][1])
            x3=(detections['detection_boxes'][j][2])
            y3=(detections['detection_boxes'][j][3])
            x4=max(x0,x2)
            y4=max(y0,y2)
            x5=min(x1,x3)
            y5=min(y1,y3)
            if x4>x5 or y4>y5:
                continue
            prevarea=(x3-x2)*(y3-y2)
            commonarea=(x5-x4)*(y5-y4)
            ins1=curarea/commonarea
            ins2=prevarea/commonarea
            ins=commonarea/(curarea+prevarea-commonarea)  
            print(ins1,end=' ')
            if(ins>=0.5):
                ok=0
                cur=detections['detection_classes'][j]
                print(classes[cur])
                break
        if ok==1:
            mark[i]=1
            cur=detections['detection_classes'][i]
            #myletters.append(classes[cur])
        print(ok)
    #verification
    for i in range(0,15):
        if mark[i]==0 or avver==0:
            continue
        if detections['detection_classes'][i]>38:
            continue
        x0=int(detections['detection_boxes'][i][0]*row)
        y0=int(detections['detection_boxes'][i][1]*col)
        x1=int(detections['detection_boxes'][i][2]*row)
        y1=int(detections['detection_boxes'][i][3]*col)
        #print(y0,y1,x0,x1)
        currImg = cpimg[x0:x1,y0:y1]
        
        curscore = detections['detection_scores'][i]
        curclass = detections['detection_classes'][i]
        label,conf = verify(currImg)
        #print(ulta[label],conf)
        #print(curclass,curscore)
        if conf>curscore and ulta[label]!=curclass and ulta[label]!=-1:
            detections['detection_classes'][i]=ulta[label]
            detections['detection_scores'][i]=conf
            
    for i in range(0,15):
        if(detections['detection_scores'][i]<0.2):
            continue
        if mark[i]==0:
            continue
        cur=detections['detection_classes'][i]
        cur=classes[cur]
        y0=(detections['detection_boxes'][i][1])
        y1=(detections['detection_boxes'][i][3])
        pair = (y0,cur,y1)
        myletters.append(pair)
    myletters.sort(key = lambda x: x[0])
    #print(myletters)
    for i in range(len(myletters)-1,-1,-1):
        y0=myletters[i][0]
        curr=myletters[i][1]
        y1=myletters[i][2]
        if curr=='ু' or curr=='্র':
            mxarea=0
            mxno=i-1
            for j in range(0,len(myletters)):
                if i==j:
                    continue
                y2=myletters[j][0]
                y3=myletters[j][2]
                curcommon = min(y3,y1)-max(y0,y2)
                if curcommon>mxarea:
                    mxarea = curcommon
                    mxno=j
            if mxno!=(i-1):
                myletters[i],myletters[i+1]=myletters[i+1],myletters[i]
    
    res_list = [x[1] for x in myletters]
    print(res_list)
    
        
    for i in range(len(res_list)-2, -1, -1):
        x=res_list[i]
        y=res_list[i+1]
        if x=='ে' or x=='ি':
            res_list[i],res_list[i+1]=res_list[i+1],res_list[i]
    for i in range(len(res_list)-2, -1, -1):
        x=res_list[i]
        y=res_list[i+1]
        print(x,y)
        if x=='অ' and y=='া':
            print('yo')
            res_list[i]='আ'
            res_list.pop(i+1)
    print(res_list)
    for i in res_list:
        print(i,end='')
        
    print(' ') 
    return res_list




# Define a flask app
# app = Flask(__name__)

# Model saved with Keras model.save()

# Load your trained model
# model = load_model(MODEL_PATH)
#model._make_predict_function()          # Necessary
# print('Model loaded. Start serving...')

# You can also use pretrained model from Keras
# Check https://keras.io/applications/
#from keras.applications.resnet50 import ResNet50
#model = ResNet50(weights='imagenet')
#model.save('')
# print('Model loaded. Check http://127.0.0.1:5000/')
avver=0
clicked=1
wp = None; bp = None;

category_index = label_map_util.create_category_index_from_labelmap(ANNOTATION_PATH+'/label_map.pbtxt')
classes=['অ','ই','উ','এ','ও','ক','খ','গ','ঘ','চ','ছ','জ','ঝ','ট','ঠ','ড','ত','থ','দ','ধ','ন','প','ফ','ব','ভ','ম','য','র','ল','শ','ষ','স','হ','ড়','য়','ৎ','ং','ঁ','০','১','২','৩','৪','৫','৭','৮','া','ি','ী','ে','ু','্র','্য']
labels=[1,2,4,7,9,11,12,13,14,16,17,18,19,21,22,23,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,45,46,47,49,50,51,52,53,54,55,57,58,60,61,62,63,64,66,67]
ulta=[0,-1,1,-1,2,-1,-1,3,-1,4,-1,5,6,7,8,-1,9,10,11,12,-1,13,14,15,-1,-1,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,-1,34,35,36,-1,37,38,39,40,41,42,43,-1,44,45,-1,46,47,48,49,50,-1,51,52]


def model_predict(word):
    #img = cv2.imread(img_path,cv2.IMREAD_GRAYSCALE)
    '''
    if clicked==1:
        bp = 66
        wp = 160
        mode = "GCMODE"
        if mode == "GCMODE":
            img= hpf(img,kSize = 51)
            wp = 127
            img = wps(img,wp)
            img = bps(img)
        elif mode == "RMODE":
            bps()
            wps()
        elif mode == "SMODE":
            bps()
            wps()
            baw()
    img = cv2.fastNlMeansDenoising(img, img, 50.0, 7, 21) 
    print("\ndone.")
    xs=img.shape
    if len(xs)==3:
        img = img[:,:,0]

    img = cv.adaptiveThreshold(img,255,cv.ADAPTIVE_THRESH_GAUSSIAN_C,cv.THRESH_BINARY,11,2)
    angeel = deskew(img)
    if angeel!=0:
        img = deskew2(img,angeel)
    ho,wo=img.shape
    area=ho*wo
    ara=words(img,25,11,7,area/5000)
    ara.reverse()
    #cv2.imshow('input image',img)
    sz=len(ara)
    for i in range(0,sz):
        ara[i]=sorted(ara[i], key=lambda entry:entry[0][0])
    cnt2=0
    files = glob.glob('Tensorflow/myimages/*')
    for f in files:
        os.remove(f)
    for i in range(0,sz):
        #print(ara[i].shape)
        tmp=ara[i]
        sz2=len(tmp)
        if i%10==0:
            cnt2=cnt2+1
        for j in range(0,sz2):
            a,b=tmp[j]
            b = cv2.adaptiveThreshold(b,255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C,cv2.THRESH_BINARY,11,2)
            if j<10:
                cnt3=0
            elif j<20:
                cnt3=1
            else:
                cnt3=2
            cv2.imwrite('Tensorflow/myimages/ocr %d%d%d%d.jpg' % (cnt2,i,cnt3,j), b)
            #cv2.imshow('Crop %d%d' % (i,j), b)
    cv2.waitKey(0)
    
    PATH_TO_TEST_IMAGES_DIR = pathlib.Path('Tensorflow/myimages')
    TEST_IMAGE_PATHS = (list(PATH_TO_TEST_IMAGES_DIR.glob("*.jpg"))+list(PATH_TO_TEST_IMAGES_DIR.glob("*.jpeg"))) #+list(PATH_TO_TEST_IMAGES_DIR.glob("*.png"))
    print(len(TEST_IMAGE_PATHS))
    final = []
    for image_path in TEST_IMAGE_PATHS:
        print("ovi")
        print(image_path)
        frame = cv2.imread(str(image_path))
        x=str(image_path)
        print(x[25])
        # gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        final.append((detect_frame(frame),x[25]))
    '''
    frame = cv2.fastNlMeansDenoising(word,word, 50.0, 7, 21) 
    xs = frame.shape
    if(len(xs)==3):
        frame = frame[:,:,0]
    frame= cv2.adaptiveThreshold(frame,255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C,cv2.THRESH_BINARY,11,2)
    frame = cv2.cvtColor(frame, cv2.COLOR_GRAY2BGR)
    # x=str(img_path)
    #print(x[25])
        # gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    image_np = np.array(frame)
    cpimg = frame.copy()
    input_tensor = tf.convert_to_tensor(np.expand_dims(image_np, 0), dtype=tf.float32)
    image_t, shapes = detection_model.preprocess(input_tensor)
    prediction_dict = detection_model.predict(image_t, shapes)
    detections = detection_model.postprocess(prediction_dict, shapes)
    # print(len(detections))
    num_detections = int(detections.pop('num_detections'))
    #print("hello")
    #print(num_detections)
    
    #print(len(detections['detection_scores']))
    detections = {key: value[0, :num_detections].numpy()
                  for key, value in detections.items()}
    detections['num_detections'] = num_detections
    row,col,dummy = image_np.shape
    # detection_classes should be ints.
    detections['detection_classes'] = detections['detection_classes'].astype(np.int64)
    #print(detections['detection_classes'])
    mark = [0]*15
    myletters = []
    for i in range(0,15):
        curi=detections['detection_classes'][i]
        classi=classes[curi]
        # print(classes[curi],end='-')
        cur=detections['detection_scores'][i]
        if(cur<0.2):
            continue
        # print(cur,end=' ')
        # print(detections['detection_boxes'][i], end=' ')
        x0=(detections['detection_boxes'][i][0])
        y0=(detections['detection_boxes'][i][1])
        x1=(detections['detection_boxes'][i][2])
        y1=(detections['detection_boxes'][i][3])
        curarea=(x1-x0)*(y1-y0)
        ok=1
        for j in range(0,i):
            #print(mark[j])
            if mark[j]==0:
                continue
            curj=detections['detection_classes'][j]
            classj=classes[curj]
            
            if classi=='ি' or classj=='ি':
                if classi!=classj:
                    continue
            if classi=='ী' or classj=='ী':
                if classi!=classj:
                    continue
            
            x2=(detections['detection_boxes'][j][0])
            y2=(detections['detection_boxes'][j][1])
            x3=(detections['detection_boxes'][j][2])
            y3=(detections['detection_boxes'][j][3])
            x4=max(x0,x2)
            y4=max(y0,y2)
            x5=min(x1,x3)
            y5=min(y1,y3)
            if x4>x5 or y4>y5:
                continue
            prevarea=(x3-x2)*(y3-y2)
            commonarea=(x5-x4)*(y5-y4)
            ins1=curarea/commonarea
            ins2=prevarea/commonarea
            ins=commonarea/(curarea+prevarea-commonarea)  
            # print(ins1,end=' ')
            if(ins>=0.5):
                ok=0
                cur=detections['detection_classes'][j]
                # print(classes[cur])
                break
        if ok==1:
            mark[i]=1
            cur=detections['detection_classes'][i]
            #myletters.append(classes[cur])
        # print(ok)
    #verification
    for i in range(0,15):
        if mark[i]==0 or avver==0:
            continue
        if detections['detection_classes'][i]>38:
            continue
        x0=int(detections['detection_boxes'][i][0]*row)
        y0=int(detections['detection_boxes'][i][1]*col)
        x1=int(detections['detection_boxes'][i][2]*row)
        y1=int(detections['detection_boxes'][i][3]*col)
        #print(y0,y1,x0,x1)
        currImg = cpimg[x0:x1,y0:y1]
        
        curscore = detections['detection_scores'][i]
        curclass = detections['detection_classes'][i]
        label,conf = verify(currImg)
        #print(ulta[label],conf)
        #print(curclass,curscore)
        if conf>curscore and ulta[label]!=curclass and ulta[label]!=-1:
            detections['detection_classes'][i]=ulta[label]
            detections['detection_scores'][i]=conf
            
    for i in range(0,15):
        if(detections['detection_scores'][i]<0.2):
            continue
        if mark[i]==0:
            continue
        cur=detections['detection_classes'][i]
        cur=classes[cur]
        y0=(detections['detection_boxes'][i][1])
        y1=(detections['detection_boxes'][i][3])
        pair = (y0,cur,y1)
        myletters.append(pair)
    myletters.sort(key = lambda x: x[0])
    #print(myletters)
    for i in range(len(myletters)-1,-1,-1):
        y0=myletters[i][0]
        curr=myletters[i][1]
        y1=myletters[i][2]
        if curr=='ু' or curr=='্র':
            mxarea=0
            mxno=i-1
            for j in range(0,len(myletters)):
                if i==j:
                    continue
                y2=myletters[j][0]
                y3=myletters[j][2]
                curcommon = min(y3,y1)-max(y0,y2)
                if curcommon>mxarea:
                    mxarea = curcommon
                    mxno=j
            if mxno!=(i-1):
                myletters[i],myletters[i+1]=myletters[i+1],myletters[i]
    
    res_list = [x[1] for x in myletters]
    # print(res_list)
    
        
    for i in range(len(res_list)-2, -1, -1):
        x=res_list[i]
        y=res_list[i+1]
        if x=='ে' or x=='ি':
            res_list[i],res_list[i+1]=res_list[i+1],res_list[i]
    for i in range(len(res_list)-2, -1, -1):
        x=res_list[i]
        y=res_list[i+1]
        # print(x,y)
        if x=='অ' and y=='া':
            # print('yo')
            res_list[i]='আ'
            res_list.pop(i+1)
    # print(res_list)
    output=''
    for i in res_list:
        output=output+i
        
    # print(' ') 
    # time_now  = datetime.datetime.now().strftime('%m_%d_%Y_%I_%M_%S_%p') 
    # # print(time_now)
    # date  = datetime.date.today().strftime('%Y_%m_%d') 
    # # print(date)
    # folderName = "created/"+date
    # if(not os.path.isdir(folderName)):
    #     os.makedirs(folderName)
    # fileName = folderName+ "/" + time_now + ".png"
    # cv2.imwrite(fileName,word)
    pil_image = Image.fromarray(word)
        #pil_image.convert("RGBA")
    for i in range(0,15):
        if mark[i]==0:
            continue
        x0=(detections['detection_boxes'][i][0])*row
        y0=(detections['detection_boxes'][i][1])*col
        x1=(detections['detection_boxes'][i][2])*row
        y1=(detections['detection_boxes'][i][3])*col
        pt1 = (y0,x0)
        pt2 = (y1,x1)
        # color = (0, 0, 255)  # Red color in BGR format
        # thickness = 2  # Border thickness in pixels
        # word = cv2.rectangle(word, pt1, pt2, color, thickness)
        draw = ImageDraw.Draw(pil_image,"RGBA")
        curi=detections['detection_classes'][i]
        classi=classes[curi]
        shape = [(y0,x0), (y1, x1)]
        draw.rectangle(shape,fill=(0, 100, 200, 127))
        draw.rectangle(shape, outline=(0, 0, 0, 127), width=3)
        bbox = draw.textbbox(pt1, classi, font=font)
        draw.rectangle(bbox, fill=(200, 100, 0, 200))
        draw.text(pt1, classi, font=font, fill=(0,0,0,255))
    newWordImg = np.asarray(pil_image)
    return output, newWordImg
    '''
    output=''
    for i in range(0,len(final)):
        ara=final[i][0]
        numb=final[i][1]
        if i>0 and numb!=final[i-1][1]:
            output= output+'\n'
        word = ''.join(ara)
        #corrected_word = get_campaign(word)
        output= output + word
        #print(corrected_word,end='')
        output = output + ' '
    return output
    '''

import gradio as gr
HF_TOKEN = os.getenv("SECRET_TOKEN")
hf_writer = gr.HuggingFaceDatasetSaver(HF_TOKEN, "word-flag-data")

demo = gr.Interface(fn=model_predict, inputs= "paint", outputs=["text","image"],
    # title= "Bangla Word OCR",
	description="Reduce pen ink size from Pen Icon(🖋️) for better results!",
    # examples=[
    #     ["Tensorflow/workspace/images/tmpbvc06xxf.png"],
    #     ["Tensorflow/workspace/images/tmpfhin6fzg.png"],
    #     ["Tensorflow/workspace/images/tmprhqli3yl.png"],
    # ],
    article=citation_text,allow_flagging="auto",flagging_callback=hf_writer)
demo.launch()