dataset.py

import os
import sys
import re
import six
import math
import lmdb
import random
import torch

from natsort import natsorted
from PIL import Image
import numpy as np
from torch.utils.data import Dataset, ConcatDataset, Subset
from torch._utils import _accumulate
import torchvision.transforms as transforms
from utils import SynthGenerator

import phoc
import pdb

class Batch_Balanced_Dataset(object):

    def __init__(self, opt):
        """
        Modulate the data ratio in the batch.
        For example, when select_data is "MJ-ST" and batch_ratio is "0.5-0.5",
        the 50% of the batch is filled with MJ and the other 50% of the batch is filled with ST.
        """
        log = open(os.path.join(opt.exp_dir,opt.exp_name,'log_dataset.txt'), 'a')
        dashed_line = '-' * 80
        print(dashed_line)
        log.write(dashed_line + '\n')
        print(f'dataset_root: {opt.train_data}\nopt.select_data: {opt.select_data}\nopt.batch_ratio: {opt.batch_ratio}')
        log.write(f'dataset_root: {opt.train_data}\nopt.select_data: {opt.select_data}\nopt.batch_ratio: {opt.batch_ratio}\n')
        assert len(opt.select_data) == len(opt.batch_ratio)

        _AlignCollate = AlignCollate(imgH=opt.imgH, imgW=opt.imgW, keep_ratio_with_pad=opt.PAD)
        self.data_loader_list = []
        self.dataloader_iter_list = []
        batch_size_list = []
        Total_batch_size = 0
        for selected_d, batch_ratio_d in zip(opt.select_data, opt.batch_ratio):
            _batch_size = max(round(opt.batch_size * float(batch_ratio_d)), 1)
            print(dashed_line)
            log.write(dashed_line + '\n')
            _dataset, _dataset_log = hierarchical_dataset(root=opt.train_data, opt=opt, select_data=[selected_d])
            total_number_dataset = len(_dataset)
            log.write(_dataset_log)

            """
            The total number of data can be modified with opt.total_data_usage_ratio.
            ex) opt.total_data_usage_ratio = 1 indicates 100% usage, and 0.2 indicates 20% usage.
            See 4.2 section in our paper.
            """
            number_dataset = int(total_number_dataset * float(opt.total_data_usage_ratio))
            dataset_split = [number_dataset, total_number_dataset - number_dataset]
            indices = range(total_number_dataset)
            _dataset, _ = [Subset(_dataset, indices[offset - length:offset])
                           for offset, length in zip(_accumulate(dataset_split), dataset_split)]
            selected_d_log = f'num total samples of {selected_d}: {total_number_dataset} x {opt.total_data_usage_ratio} (total_data_usage_ratio) = {len(_dataset)}\n'
            selected_d_log += f'num samples of {selected_d} per batch: {opt.batch_size} x {float(batch_ratio_d)} (batch_ratio) = {_batch_size}'
            print(selected_d_log)
            log.write(selected_d_log + '\n')
            batch_size_list.append(str(_batch_size))
            Total_batch_size += _batch_size

            _data_loader = torch.utils.data.DataLoader(
                _dataset, batch_size=_batch_size,
                shuffle=True,
                num_workers=int(opt.workers),
                collate_fn=_AlignCollate, pin_memory=True)
            self.data_loader_list.append(_data_loader)
            self.dataloader_iter_list.append(iter(_data_loader))

        Total_batch_size_log = f'{dashed_line}\n'
        batch_size_sum = '+'.join(batch_size_list)
        Total_batch_size_log += f'Total_batch_size: {batch_size_sum} = {Total_batch_size}\n'
        Total_batch_size_log += f'{dashed_line}'
        opt.batch_size = Total_batch_size

        print(Total_batch_size_log)
        log.write(Total_batch_size_log + '\n')
        log.close()
        
        self.pairText = opt.pairText
        self.lexicons=[]
        out_of_char = f'[^{opt.character}]'
        if opt.pairText == True:
            #read lexicons file
            with open(opt.lexFile,'r') as lexF:
                for line in lexF:
                    lexWord = line[:-1]
                    if opt.fixedString and len(lexWord)!=opt.batch_exact_length:
                        continue
                    if len(lexWord) <= opt.batch_max_length and not(re.search(out_of_char, lexWord.lower())) and len(lexWord) >= opt.batch_min_length:
                        self.lexicons.append(lexWord)
            

    def get_batch(self):
        
        balanced_batch_images = []
        balanced_batch_texts = []

        for i, data_loader_iter in enumerate(self.dataloader_iter_list):
            try:
                
                image, text = data_loader_iter.next()
                
                balanced_batch_images.append(image)
                balanced_batch_texts += text
            except StopIteration:
                self.dataloader_iter_list[i] = iter(self.data_loader_list[i])
                image, text = self.dataloader_iter_list[i].next()
                balanced_batch_images.append(image)
                balanced_batch_texts += text
            except ValueError:
                pass

        balanced_batch_images = torch.cat(balanced_batch_images, 0)

        if self.pairText:
            return balanced_batch_images, balanced_batch_texts, random.sample(self.lexicons,len(balanced_batch_texts))
        else:
            return balanced_batch_images, balanced_batch_texts


def hierarchical_dataset(root, opt, select_data='/'):
    """ select_data='/' contains all sub-directory of root directory """
    dataset_list = []
    dataset_log = f'dataset_root:    {root}\t dataset: {select_data[0]}'
    print(dataset_log)
    dataset_log += '\n'
    for dirpath, dirnames, filenames in os.walk(root+'/'):
        if not dirnames:
            select_flag = False
            for selected_d in select_data:
                if selected_d in dirpath:
                    select_flag = True
                    break

            if select_flag:
                if opt.style_input:
                    if opt.style_content_input:
                        dataset = LmdbStyleContentDataset(dirpath, opt)
                    else:
                        dataset = LmdbStyleDataset(dirpath, opt)
                else:
                    dataset = LmdbDataset(dirpath, opt)
                sub_dataset_log = f'sub-directory:\t/{os.path.relpath(dirpath, root)}\t num samples: {len(dataset)}'
                print(sub_dataset_log)
                dataset_log += f'{sub_dataset_log}\n'
                dataset_list.append(dataset)

    concatenated_dataset = ConcatDataset(dataset_list)

    return concatenated_dataset, dataset_log


class LmdbDataset(Dataset):

    def __init__(self, root, opt):

        self.root = root
        self.opt = opt
        self.env = lmdb.open(root, max_readers=32, readonly=True, lock=False, readahead=False, meminit=False)
        if not self.env:
            print('cannot create lmdb from %s' % (root))
            sys.exit(0)

        with self.env.begin(write=False) as txn:
            nSamples = int(txn.get('num-samples'.encode()))
            self.nSamples = nSamples
            print('nSamples:::::::::::::',nSamples)

            if self.opt.data_filtering_off:
                # for fast check or benchmark evaluation with no filtering
                self.filtered_index_list = [index + 1 for index in range(self.nSamples)]
            else:
                """ Filtering part
                If you want to evaluate IC15-2077 & CUTE datasets which have special character labels,
                use --data_filtering_off and only evaluate on alphabets and digits.
                see https://github.com/clovaai/deep-text-recognition-benchmark/blob/6593928855fb7abb999a99f428b3e4477d4ae356/dataset.py#L190-L192

                And if you want to evaluate them with the model trained with --sensitive option,
                use --sensitive and --data_filtering_off,
                see https://github.com/clovaai/deep-text-recognition-benchmark/blob/dff844874dbe9e0ec8c5a52a7bd08c7f20afe704/test.py#L137-L144
                """
                self.filtered_index_list = []
                for index in range(self.nSamples):
                    index += 1  # lmdb starts with 1
                    label_key = 'label-%09d'.encode() % index
                    label = txn.get(label_key).decode('utf-8')

                    if self.opt.fixedString and len(label) != self.opt.batch_exact_length :
                        continue
                    elif len(label) > self.opt.batch_max_length or len(label)<self.opt.batch_min_length:
                        # print(f'The length of the label is longer than max_length: length
                        # {len(label)}, {label} in dataset {self.root}')
                        continue

                    # By default, images containing characters which are not in opt.character are filtered.
                    # You can add [UNK] token to `opt.character` in utils.py instead of this filtering.
                    out_of_char = f'[^{self.opt.character}]'
                    if re.search(out_of_char, label.lower()):
                        continue
                    
                    self.filtered_index_list.append(index)

                self.nSamples = len(self.filtered_index_list)

    def __len__(self):
        return self.nSamples

    def __getitem__(self, index):
        assert index <= len(self), 'index range error'
        index = self.filtered_index_list[index]

        with self.env.begin(write=False) as txn:
            label_key = 'label-%09d'.encode() % index
            
            label = txn.get(label_key).decode('utf-8')
            
            
            img_key = 'image-%09d'.encode() % index
            imgbuf = txn.get(img_key)

            buf = six.BytesIO()
            buf.write(imgbuf)
            buf.seek(0)
            try:
                if self.opt.rgb:
                    img = Image.open(buf).convert('RGB')  # for color image
                else:
                    img = Image.open(buf).convert('L')

            except IOError:
                print(f'Corrupted image for {index}')
                # make dummy image and dummy label for corrupted image.
                if self.opt.rgb:
                    img = Image.new('RGB', (self.opt.imgW, self.opt.imgH))
                else:
                    img = Image.new('L', (self.opt.imgW, self.opt.imgH))
                label = '[dummy_label]'

            if not self.opt.sensitive:
                label = label.lower()

            # We only train and evaluate on alphanumerics (or pre-defined character set in train.py)
            out_of_char = f'[^{self.opt.character}]'
            label = re.sub(out_of_char, '', label)

        return (img, label)


class LmdbStyleDataset(Dataset):

    def __init__(self, root, opt):

        self.root = root
        self.opt = opt
        self.env = lmdb.open(root, max_readers=32, readonly=True, lock=False, readahead=False, meminit=False)
        if not self.env:
            print('cannot create lmdb from %s' % (root))
            sys.exit(0)

        with self.env.begin(write=False) as txn:
            
            nSamples = int(txn.get('num-samples'.encode()))
            self.nSamples = nSamples
            print('nSamples:::::::::::::',nSamples)

            if self.opt.data_filtering_off:
                # for fast check or benchmark evaluation with no filtering
                self.filtered_index_list = [index + 1 for index in range(self.nSamples)]
            else:
                """ Filtering part
                If you want to evaluate IC15-2077 & CUTE datasets which have special character labels,
                use --data_filtering_off and only evaluate on alphabets and digits.
                see https://github.com/clovaai/deep-text-recognition-benchmark/blob/6593928855fb7abb999a99f428b3e4477d4ae356/dataset.py#L190-L192

                And if you want to evaluate them with the model trained with --sensitive option,
                use --sensitive and --data_filtering_off,
                see https://github.com/clovaai/deep-text-recognition-benchmark/blob/dff844874dbe9e0ec8c5a52a7bd08c7f20afe704/test.py#L137-L144
                """
                self.filtered_index_list = []
                for index in range(self.nSamples):
                    index += 1  # lmdb starts with 1
                    label1_key = 'label1-%09d'.encode() % index
                    label1 = txn.get(label1_key).decode('utf-8')

                    label2_key = 'label2-%09d'.encode() % index
                    label2 = txn.get(label2_key).decode('utf-8')
                    
                    if self.opt.fixedString and (len(label1) != self.opt.batch_exact_length or len(label2) != self.opt.batch_exact_length) :
                        continue
                    elif len(label1) > self.opt.batch_max_length or len(label1)<self.opt.batch_min_length or len(label2) > self.opt.batch_max_length or len(label2)<self.opt.batch_min_length:
                        # print(f'The length of the label is longer than max_length: length
                        # {len(label)}, {label} in dataset {self.root}')
                        continue

                    # By default, images containing characters which are not in opt.character are filtered.
                    # You can add [UNK] token to `opt.character` in utils.py instead of this filtering.
                    out_of_char = f'[^{self.opt.character}]'
                    if re.search(out_of_char, label1.lower()) or re.search(out_of_char, label2.lower()):
                        continue
                    
                    self.filtered_index_list.append(index)

                self.nSamples = len(self.filtered_index_list)

    def __len__(self):
        return self.nSamples

    def __getitem__(self, index):
        assert index <= len(self), 'index range error'
        index = self.filtered_index_list[index]

        with self.env.begin(write=False) as txn:
            label1_key = 'label1-%09d'.encode() % index
            label2_key = 'label2-%09d'.encode() % index
            
            label1 = txn.get(label1_key).decode('utf-8')
            label2 = txn.get(label2_key).decode('utf-8')
            
            
            img1_key = 'image1-%09d'.encode() % index
            img2_key = 'image2-%09d'.encode() % index
            img1buf = txn.get(img1_key)
            img2buf = txn.get(img2_key)

            buf1 = six.BytesIO()
            buf1.write(img1buf)
            buf1.seek(0)

            buf2 = six.BytesIO()
            buf2.write(img2buf)
            buf2.seek(0)

            try:
                if self.opt.rgb:
                    img1 = Image.open(buf1).convert('RGB')  # for color image
                    img2 = Image.open(buf2).convert('RGB')  # for color image
                else:
                    img1 = Image.open(buf1).convert('L')
                    img2 = Image.open(buf2).convert('L')

            except IOError:
                print(f'Corrupted image for {index}')
                # make dummy image and dummy label for corrupted image.
                if self.opt.rgb:
                    img1 = Image.new('RGB', (self.opt.imgW, self.opt.imgH))
                    img2 = Image.new('RGB', (self.opt.imgW, self.opt.imgH))
                else:
                    img1 = Image.new('L', (self.opt.imgW, self.opt.imgH))
                    img2 = Image.new('L', (self.opt.imgW, self.opt.imgH))
                label1 = '[dummy_label]'
                label2 = '[dummy_label]'

            if not self.opt.sensitive:
                label1 = label1.lower()
                label2 = label2.lower()

            # We only train and evaluate on alphanumerics (or pre-defined character set in train.py)
            out_of_char = f'[^{self.opt.character}]'
            label1 = re.sub(out_of_char, '', label1)
            label2 = re.sub(out_of_char, '', label2)

        return (img1, img2, label1, label2)

class LmdbStylePHOCDataset(Dataset):

    def __init__(self, root, opt):

        self.root = root
        self.opt = opt
        self.env = lmdb.open(root, max_readers=32, readonly=True, lock=False, readahead=False, meminit=False)
        self.phocObj = phoc_gen(opt)

        if not self.env:
            print('cannot create lmdb from %s' % (root))
            sys.exit(0)

        with self.env.begin(write=False) as txn:
            
            nSamples = int(txn.get('num-samples'.encode()))
            self.nSamples = nSamples
            print('nSamples:::::::::::::',nSamples)

            if self.opt.data_filtering_off:
                # for fast check or benchmark evaluation with no filtering
                self.filtered_index_list = [index + 1 for index in range(self.nSamples)]
            else:
                """ Filtering part
                If you want to evaluate IC15-2077 & CUTE datasets which have special character labels,
                use --data_filtering_off and only evaluate on alphabets and digits.
                see https://github.com/clovaai/deep-text-recognition-benchmark/blob/6593928855fb7abb999a99f428b3e4477d4ae356/dataset.py#L190-L192

                And if you want to evaluate them with the model trained with --sensitive option,
                use --sensitive and --data_filtering_off,
                see https://github.com/clovaai/deep-text-recognition-benchmark/blob/dff844874dbe9e0ec8c5a52a7bd08c7f20afe704/test.py#L137-L144
                """
                self.filtered_index_list = []
                for index in range(self.nSamples):
                    index += 1  # lmdb starts with 1
                    label1_key = 'label1-%09d'.encode() % index
                    label1 = txn.get(label1_key).decode('utf-8')

                    label2_key = 'label2-%09d'.encode() % index
                    label2 = txn.get(label2_key).decode('utf-8')
                    
                    if self.opt.fixedString and (len(label1) != self.opt.batch_exact_length or len(label2) != self.opt.batch_exact_length) :
                        continue
                    elif len(label1) > self.opt.batch_max_length or len(label1)<self.opt.batch_min_length or len(label2) > self.opt.batch_max_length or len(label2)<self.opt.batch_min_length:
                        # print(f'The length of the label is longer than max_length: length
                        # {len(label)}, {label} in dataset {self.root}')
                        continue

                    # By default, images containing characters which are not in opt.character are filtered.
                    # You can add [UNK] token to `opt.character` in utils.py instead of this filtering.
                    out_of_char = f'[^{self.opt.character}]'
                    if re.search(out_of_char, label1.lower()) or re.search(out_of_char, label2.lower()):
                        continue
                    
                    self.filtered_index_list.append(index)

                self.nSamples = len(self.filtered_index_list)

    def __len__(self):
        return self.nSamples

    def __getitem__(self, index):
        assert index <= len(self), 'index range error'
        index = self.filtered_index_list[index]

        with self.env.begin(write=False) as txn:
            label1_key = 'label1-%09d'.encode() % index
            label2_key = 'label2-%09d'.encode() % index
            
            label1 = txn.get(label1_key).decode('utf-8')
            label2 = txn.get(label2_key).decode('utf-8')
            
            
            img1_key = 'image1-%09d'.encode() % index
            img2_key = 'image2-%09d'.encode() % index
            img1buf = txn.get(img1_key)
            img2buf = txn.get(img2_key)

            buf1 = six.BytesIO()
            buf1.write(img1buf)
            buf1.seek(0)

            buf2 = six.BytesIO()
            buf2.write(img2buf)
            buf2.seek(0)

            try:
                if self.opt.rgb:
                    img1 = Image.open(buf1).convert('RGB')  # for color image
                    img2 = Image.open(buf2).convert('RGB')  # for color image
                else:
                    img1 = Image.open(buf1).convert('L')
                    img2 = Image.open(buf2).convert('L')

            except IOError:
                print(f'Corrupted image for {index}')
                # make dummy image and dummy label for corrupted image.
                if self.opt.rgb:
                    img1 = Image.new('RGB', (self.opt.imgW, self.opt.imgH))
                    img2 = Image.new('RGB', (self.opt.imgW, self.opt.imgH))
                else:
                    img1 = Image.new('L', (self.opt.imgW, self.opt.imgH))
                    img2 = Image.new('L', (self.opt.imgW, self.opt.imgH))
                label1 = '[dummy_label]'
                label2 = '[dummy_label]'

            if not self.opt.sensitive:
                label1 = label1.lower()
                label2 = label2.lower()

            # We only train and evaluate on alphanumerics (or pre-defined character set in train.py)
            out_of_char = f'[^{self.opt.character}]'
            label1 = re.sub(out_of_char, '', label1)
            label2 = re.sub(out_of_char, '', label2)
            
        return (img1, img2, label1, label2, self.phocObj.getPhoc(label1), self.phocObj.getPhoc(label2))

class LmdbStyleContentDataset(Dataset):

    def __init__(self, root, opt, dataMode=False):

        self.root = root
        self.opt = opt
        self.env = lmdb.open(root, max_readers=32, readonly=True, lock=False, readahead=False, meminit=False)
        
        self.synthGen = SynthGenerator('/private/home/pkrishnan/fonts/fonts-10-path.txt',imgSize=(64,256))

        if not self.env:
            print('cannot create lmdb from %s' % (root))
            sys.exit(0)

        with self.env.begin(write=False) as txn:
            
            nSamples = int(txn.get('num-samples'.encode()))
            self.nSamples = nSamples
            print('nSamples:::::::::::::',nSamples)

            if self.opt.data_filtering_off:
                # for fast check or benchmark evaluation with no filtering
                self.filtered_index_list = [index + 1 for index in range(self.nSamples)]
            else:
                """ Filtering part
                If you want to evaluate IC15-2077 & CUTE datasets which have special character labels,
                use --data_filtering_off and only evaluate on alphabets and digits.
                see https://github.com/clovaai/deep-text-recognition-benchmark/blob/6593928855fb7abb999a99f428b3e4477d4ae356/dataset.py#L190-L192

                And if you want to evaluate them with the model trained with --sensitive option,
                use --sensitive and --data_filtering_off,
                see https://github.com/clovaai/deep-text-recognition-benchmark/blob/dff844874dbe9e0ec8c5a52a7bd08c7f20afe704/test.py#L137-L144
                """
                self.filtered_index_list = []
                for index in range(self.nSamples):
                    index += 1  # lmdb starts with 1
                    if dataMode:
                        #Todo: remove duplication; simplify
                        label1_key = 'label-%09d'.encode() % index
                        label1 = txn.get(label1_key).decode('utf-8')

                        label2_key = 'label-%09d'.encode() % index
                        label2 = txn.get(label2_key).decode('utf-8')
                    else:    
                        label1_key = 'label1-%09d'.encode() % index
                        label1 = txn.get(label1_key).decode('utf-8')

                        label2_key = 'label2-%09d'.encode() % index
                        label2 = txn.get(label2_key).decode('utf-8')
                    
                    if self.opt.fixedString and (len(label1) != self.opt.batch_exact_length or len(label2) != self.opt.batch_exact_length) :
                        continue
                    elif len(label1) > self.opt.batch_max_length or len(label1)<self.opt.batch_min_length or len(label2) > self.opt.batch_max_length or len(label2)<self.opt.batch_min_length:
                        # print(f'The length of the label is longer than max_length: length
                        # {len(label)}, {label} in dataset {self.root}')
                        continue

                    # By default, images containing characters which are not in opt.character are filtered.
                    # You can add [UNK] token to `opt.character` in utils.py instead of this filtering.
                    out_of_char = f'[^{self.opt.character}]'
                    if re.search(out_of_char, label1.lower()) or re.search(out_of_char, label2.lower()):
                        continue
                    
                    self.filtered_index_list.append(index)

                self.nSamples = len(self.filtered_index_list)
                self.realData = dataMode

    def __len__(self):
        return self.nSamples

    def __getitem__(self, index):
        assert index <= len(self), 'index range error'
        index = self.filtered_index_list[index]

        with self.env.begin(write=False) as txn:
            if self.realData:
                #todo: remove duplication; simplify
                label1_key = 'label-%09d'.encode() % index
                label2_key = 'label-%09d'.encode() % index
            else:
                label1_key = 'label1-%09d'.encode() % index
                label2_key = 'label2-%09d'.encode() % index
            
            label1 = txn.get(label1_key).decode('utf-8')
            label2 = txn.get(label2_key).decode('utf-8')
            
            if self.realData:
                #todo: remove duplication; simplify
                img1_key = 'image-%09d'.encode() % index
                img2_key = 'image-%09d'.encode() % index
            else:
                img1_key = 'image1-%09d'.encode() % index
                img2_key = 'image2-%09d'.encode() % index

            img1buf = txn.get(img1_key)
            img2buf = txn.get(img2_key)

            buf1 = six.BytesIO()
            buf1.write(img1buf)
            buf1.seek(0)

            buf2 = six.BytesIO()
            buf2.write(img2buf)
            buf2.seek(0)

            try:
                if self.opt.rgb:
                    img1 = Image.open(buf1).convert('RGB')  # for color image
                    img2 = Image.open(buf2).convert('RGB')  # for color image
                else:
                    img1 = Image.open(buf1).convert('L')
                    img2 = Image.open(buf2).convert('L')

            except IOError:
                print(f'Corrupted image for {index}')
                # make dummy image and dummy label for corrupted image.
                if self.opt.rgb:
                    img1 = Image.new('RGB', (self.opt.imgW, self.opt.imgH))
                    img2 = Image.new('RGB', (self.opt.imgW, self.opt.imgH))
                else:
                    img1 = Image.new('L', (self.opt.imgW, self.opt.imgH))
                    img2 = Image.new('L', (self.opt.imgW, self.opt.imgH))
                label1 = '[dummy_label]'
                label2 = '[dummy_label]'

            if not self.opt.sensitive:
                label1 = label1.lower()
                label2 = label2.lower()

            # We only train and evaluate on alphanumerics (or pre-defined character set in train.py)
            out_of_char = f'[^{self.opt.character}]'
            label1 = re.sub(out_of_char, '', label1)
            label2 = re.sub(out_of_char, '', label2)
            label1SynthImg = self.synthGen.synthesizeWordImage(label1, 0)
            label2SynthImg = self.synthGen.synthesizeWordImage(label2, 0)
            

        return (img1, img2, label1, label2, label1SynthImg, label2SynthImg)



class LmdbTestStyleContentDataset(Dataset):

    def __init__(self, root, opt, dataMode=False):

        self.root = root
        self.opt = opt
        self.env = lmdb.open(root, max_readers=32, readonly=True, lock=False, readahead=False, meminit=False)
        
        self.synthGen = SynthGenerator('/private/home/pkrishnan/fonts/fonts-10-path.txt',imgSize=(64,256))

        if not self.env:
            print('cannot create lmdb from %s' % (root))
            sys.exit(0)

        self.pairId = []
        self.pairText = []
        if opt.outPairFile!="":
            with open(opt.outPairFile,'r') as pID:
                lines = pID.readlines()

                for currline in lines:
                    tokens = currline[:-1].split(" ")
                    labelTok = tokens[0].split('-')
                    self.pairId.append(labelTok[1])
                    self.pairText.append(tokens[2])

        with self.env.begin(write=False) as txn:
            
            nSamples = int(txn.get('num-samples'.encode()))
            self.nSamples = nSamples
            print('nSamples:::::::::::::',nSamples)

            if self.opt.data_filtering_off:
                # for fast check or benchmark evaluation with no filtering
                self.filtered_index_list = [index + 1 for index in range(self.nSamples)]
            else:
                """ Filtering part
                If you want to evaluate IC15-2077 & CUTE datasets which have special character labels,
                use --data_filtering_off and only evaluate on alphabets and digits.
                see https://github.com/clovaai/deep-text-recognition-benchmark/blob/6593928855fb7abb999a99f428b3e4477d4ae356/dataset.py#L190-L192

                And if you want to evaluate them with the model trained with --sensitive option,
                use --sensitive and --data_filtering_off,
                see https://github.com/clovaai/deep-text-recognition-benchmark/blob/dff844874dbe9e0ec8c5a52a7bd08c7f20afe704/test.py#L137-L144
                """
                self.filtered_index_list = []
                for index in range(self.nSamples):
                    index += 1  # lmdb starts with 1
                    if dataMode:
                        #Todo: remove duplication; simplify
                        label1_key = 'label-%09d'.encode() % index
                        label1 = txn.get(label1_key).decode('utf-8')

                        label2_key = 'label-%09d'.encode() % index
                        label2 = txn.get(label2_key).decode('utf-8')
                    else:    
                        label1_key = 'label1-%09d'.encode() % index
                        label1 = txn.get(label1_key).decode('utf-8')

                        label2_key = 'label2-%09d'.encode() % index
                        label2 = txn.get(label2_key).decode('utf-8')
                    
                    #read image and filter
                    if dataMode:
                        img1_key = 'image-%09d'.encode() % index
                        img1buf = txn.get(img1_key)

                        buf1 = six.BytesIO()
                        buf1.write(img1buf)
                        buf1.seek(0)

                        try:
                            if self.opt.rgb:
                                img1 = Image.open(buf1).convert('RGB')  # for color image
                            else:
                                img1 = Image.open(buf1).convert('L')
                            
                            if opt.sizeFilt and (img1.size[1]<opt.imgH_filt or (img1.size[0] < 1.25*img1.size[1])):
                                continue
                        except IOError:
                            print(f'Corrupted image for {index}')
                            continue


                    if self.opt.fixedString and (len(label1) != self.opt.batch_exact_length or len(label2) != self.opt.batch_exact_length) :
                        continue
                    elif len(label1) > self.opt.batch_max_length or len(label1)<self.opt.batch_min_length or len(label2) > self.opt.batch_max_length or len(label2)<self.opt.batch_min_length:
                        # print(f'The length of the label is longer than max_length: length
                        # {len(label)}, {label} in dataset {self.root}')
                        continue

                    # By default, images containing characters which are not in opt.character are filtered.
                    # You can add [UNK] token to `opt.character` in utils.py instead of this filtering.
                    out_of_char = f'[^{self.opt.character}]'
                    if re.search(out_of_char, label1.lower()) or re.search(out_of_char, label2.lower()):
                        continue
                    
                    self.filtered_index_list.append(index)

                self.nSamples = len(self.filtered_index_list)
                if len(self.pairId)>0:
                    print('Filtered nSamples:::::::::::::',len(self.pairId))
                else:
                    print('Filtered nSamples:::::::::::::',self.nSamples)
                self.realData = dataMode

    def __len__(self):
        if len(self.pairId)>0:
            return len(self.pairId)
        else:
            return self.nSamples

    def __getitem__(self, index):
        assert index <= len(self), 'index range error'

        if len(self.pairId)==0:
            index = self.filtered_index_list[index]

        with self.env.begin(write=False) as txn:
            
            if len(self.pairId)>0:
                if self.realData:
                    label1_key = ('label-'+self.pairId[index]).encode()
                    label2_key = ('label-'+self.pairId[index]).encode()
                else:
                    label1_key = ('label1-'+self.pairId[index]).encode()
                    label2_key = ('label2-'+self.pairId[index]).encode()
            else:
                if self.realData:
                    #todo: remove duplication; simplify
                    label1_key = 'label-%09d'.encode() % index
                    label2_key = 'label-%09d'.encode() % index
                else:
                    label1_key = 'label1-%09d'.encode() % index
                    label2_key = 'label2-%09d'.encode() % index
            
            if len(self.pairId)>0:
                label1 = txn.get(label1_key).decode('utf-8')
                label2 = self.pairText[index]
            else:
                label1 = txn.get(label1_key).decode('utf-8')
                label2 = txn.get(label2_key).decode('utf-8')

            if len(self.pairId)>0:
                if self.realData:
                    #todo: remove duplication; simplify
                    img1_key = ('image-'+self.pairId[index]).encode()
                    img2_key = ('image-'+self.pairId[index]).encode()
                else:
                    img1_key = ('image1-'+self.pairId[index]).encode()
                    img2_key = ('image2-'+self.pairId[index]).encode()
            else:
                if self.realData:
                    #todo: remove duplication; simplify
                    img1_key = 'image-%09d'.encode() % index
                    img2_key = 'image-%09d'.encode() % index
                else:
                    img1_key = 'image1-%09d'.encode() % index
                    img2_key = 'image2-%09d'.encode() % index

            img1buf = txn.get(img1_key)
            img2buf = txn.get(img2_key)

            buf1 = six.BytesIO()
            buf1.write(img1buf)
            buf1.seek(0)

            buf2 = six.BytesIO()
            buf2.write(img2buf)
            buf2.seek(0)

            try:
                if self.opt.rgb:
                    img1 = Image.open(buf1).convert('RGB')  # for color image
                    img2 = Image.open(buf2).convert('RGB')  # for color image
                else:
                    img1 = Image.open(buf1).convert('L')
                    img2 = Image.open(buf2).convert('L')

            except IOError:
                print(f'Corrupted image for {index}')
                # make dummy image and dummy label for corrupted image.
                if self.opt.rgb:
                    img1 = Image.new('RGB', (self.opt.imgW, self.opt.imgH))
                    img2 = Image.new('RGB', (self.opt.imgW, self.opt.imgH))
                else:
                    img1 = Image.new('L', (self.opt.imgW, self.opt.imgH))
                    img2 = Image.new('L', (self.opt.imgW, self.opt.imgH))
                label1 = '[dummy_label]'
                label2 = '[dummy_label]'

            if not self.opt.sensitive:
                label1 = label1.lower()
                label2 = label2.lower()

            # We only train and evaluate on alphanumerics (or pre-defined character set in train.py)
            out_of_char = f'[^{self.opt.character}]'
            label1 = re.sub(out_of_char, '', label1)
            label2 = re.sub(out_of_char, '', label2)
            label1SynthImg = self.synthGen.synthesizeWordImage(label1, 0)
            label2SynthImg = self.synthGen.synthesizeWordImage(label2, 0)
            

        return (img1, img2, label1, label2, label1SynthImg, label2SynthImg)


class RawDataset(Dataset):

    def __init__(self, root, opt):
        self.opt = opt
        self.image_path_list = []
        for dirpath, dirnames, filenames in os.walk(root):
            for name in filenames:
                _, ext = os.path.splitext(name)
                ext = ext.lower()
                if ext == '.jpg' or ext == '.jpeg' or ext == '.png':
                    self.image_path_list.append(os.path.join(dirpath, name))

        self.image_path_list = natsorted(self.image_path_list)
        self.nSamples = len(self.image_path_list)

    def __len__(self):
        return self.nSamples

    def __getitem__(self, index):

        try:
            if self.opt.rgb:
                img = Image.open(self.image_path_list[index]).convert('RGB')  # for color image
            else:
                img = Image.open(self.image_path_list[index]).convert('L')

        except IOError:
            print(f'Corrupted image for {index}')
            # make dummy image and dummy label for corrupted image.
            if self.opt.rgb:
                img = Image.new('RGB', (self.opt.imgW, self.opt.imgH))
            else:
                img = Image.new('L', (self.opt.imgW, self.opt.imgH))

        return (img, self.image_path_list[index])


class ResizeNormalize(object):

    def __init__(self, size, interpolation=Image.BICUBIC):
        self.size = size
        self.interpolation = interpolation
        self.toTensor = transforms.ToTensor()

    def __call__(self, img):
        img = img.resize(self.size, self.interpolation)
        img = self.toTensor(img)
        img.sub_(0.5).div_(0.5)
        return img

class ResizeNormalize_translate(object):

    def __init__(self, size, interpolation=Image.BICUBIC):
        self.size = size
        self.interpolation = interpolation
        self.toTensor = transforms.ToTensor()
        self.randomcrop=transforms.RandomResizedCrop((size[1],size[0]), scale=(0.99, 1.0), ratio=(1.0, 1.0), interpolation=interpolation)
        self.affine = transforms.RandomAffine(0, translate=(0.1,0.2), scale=(0.8,1.0))

    def __call__(self, img):
        # img = img.resize(self.size, self.interpolation)
        # img = self.affine(img)
        img = self.randomcrop(img)
        img = self.toTensor(img)
        img.sub_(0.5).div_(0.5)
        return img

class ResizeNormalize_Mean(object):

    def __init__(self, size, interpolation=Image.BICUBIC):
        self.size = size
        self.interpolation = interpolation
        self.toTensor = transforms.ToTensor()
        self.normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
                                 std=[0.229, 0.224, 0.225])

    def __call__(self, img):
        img = img.resize(self.size, self.interpolation)
        img = self.toTensor(img)
        img = self.normalize(img)
        # img.sub_(0.5).div_(0.5)
        return img


class NormalizePAD(object):

    def __init__(self, max_size, PAD_type='right'):
        self.toTensor = transforms.ToTensor()
        self.max_size = max_size
        self.max_width_half = math.floor(max_size[2] / 2)
        self.PAD_type = PAD_type

    def __call__(self, img):
        img = self.toTensor(img)
        img.sub_(0.5).div_(0.5)
        c, h, w = img.size()
        Pad_img = torch.FloatTensor(*self.max_size).fill_(0)
        Pad_img[:, :, :w] = img  # right pad
        
        if self.max_size[2] != w:  # add border Pad
            # Pad_img[:, :, w:] = img[:, :, w - 1].unsqueeze(2).expand(c, h, self.max_size[2] - w)
            zero_vec = torch.zeros((img.size()[0],img.size()[1]))
            Pad_img[:, :, w:] = zero_vec.unsqueeze(2).expand(c, h, self.max_size[2] - w)
        return Pad_img

class NormalizePAD_Mean(object):

    def __init__(self, max_size, PAD_type='right'):
        self.toTensor = transforms.ToTensor()
        self.normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
                                 std=[0.229, 0.224, 0.225])
        self.max_size = max_size
        self.max_width_half = math.floor(max_size[2] / 2)
        self.PAD_type = PAD_type

    def __call__(self, img):
        img = self.toTensor(img)
        # img.sub_(0.5).div_(0.5)
        img = self.normalize(img)

        c, h, w = img.size()
        Pad_img = torch.FloatTensor(*self.max_size).fill_(0)
        Pad_img[:, :, :w] = img  # right pad
        
        if self.max_size[2] != w:  # add border Pad
            # Pad_img[:, :, w:] = img[:, :, w - 1].unsqueeze(2).expand(c, h, self.max_size[2] - w)
            zero_vec = torch.zeros((img.size()[0],img.size()[1]))
            Pad_img[:, :, w:] = zero_vec.unsqueeze(2).expand(c, h, self.max_size[2] - w)
        return Pad_img

class NormalizePADVarSize(object):

    def __init__(self, max_size, PAD_type='right'):
        self.toTensor = transforms.ToTensor()
        self.max_size = max_size
        self.max_width_half = math.floor(max_size[2] / 2)
        self.PAD_type = PAD_type

    def __call__(self, img):
        img = self.toTensor(img)
        img.sub_(0.5).div_(0.5)
        # c, h, w = img.size()
        # Pad_img = torch.FloatTensor(*self.max_size).fill_(0)
        # Pad_img[:, :, :w] = img  # right pad
        
        # if self.max_size[2] != w:  # add border Pad
        #     # Pad_img[:, :, w:] = img[:, :, w - 1].unsqueeze(2).expand(c, h, self.max_size[2] - w)
        #     zero_vec = torch.zeros((img.size()[0],img.size()[1]))
        #     Pad_img[:, :, w:] = zero_vec.unsqueeze(2).expand(c, h, self.max_size[2] - w)
        return img

class AlignCollate(object):

    def __init__(self, imgH=32, imgW=100, keep_ratio_with_pad=False):
        self.imgH = imgH
        self.imgW = imgW
        self.keep_ratio_with_pad = keep_ratio_with_pad

    def __call__(self, batch):
        batch = filter(lambda x: x is not None, batch)
        images, labels = zip(*batch)

        if self.keep_ratio_with_pad:  # same concept with 'Rosetta' paper
            resized_max_w = self.imgW
            input_channel = 3 if images[0].mode == 'RGB' else 1
            transform = NormalizePAD((input_channel, self.imgH, resized_max_w))

            resized_images = []
            for image in images:
                w, h = image.size
                ratio = w / float(h)
                if math.ceil(self.imgH * ratio) > self.imgW:
                    resized_w = self.imgW
                else:
                    resized_w = math.ceil(self.imgH * ratio)

                resized_image = image.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_images.append(transform(resized_image))
                # resized_image.save('./image_test/%d_test.jpg' % w)

            image_tensors = torch.cat([t.unsqueeze(0) for t in resized_images], 0)

        else:
            transform = ResizeNormalize((self.imgW, self.imgH))
            image_tensors = [transform(image) for image in images]
            image_tensors = torch.cat([t.unsqueeze(0) for t in image_tensors], 0)

        return image_tensors, labels

class AlignFontCollate(object):

    def __init__(self, imgH=32, imgW=100, keep_ratio_with_pad=False):
        self.imgH = imgH
        self.imgW = imgW
        self.keep_ratio_with_pad = keep_ratio_with_pad

    def __call__(self, batch):
        batch = filter(lambda x: x is not None, batch)
        images, labels = zip(*batch)

        if self.keep_ratio_with_pad:  # same concept with 'Rosetta' paper
            resized_max_w = self.imgW
            input_channel = 3 if images[0].mode == 'RGB' else 1
            transform = NormalizePAD((input_channel, self.imgH, resized_max_w))

            resized_images = []
            for image in images:
                w, h = image.size
                ratio = w / float(h)
                if math.ceil(self.imgH * ratio) > self.imgW:
                    resized_w = self.imgW
                else:
                    resized_w = math.ceil(self.imgH * ratio)

                resized_image = image.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_images.append(transform(resized_image))
                # resized_image.save('./image_test/%d_test.jpg' % w)

            image_tensors = torch.cat([t.unsqueeze(0) for t in resized_images], 0)

        else:
            transform = ResizeNormalize((self.imgW, self.imgH))
            image_tensors = [transform(image) for image in images]
            image_tensors = torch.cat([t.unsqueeze(0) for t in image_tensors], 0)
        
        return image_tensors, torch.LongTensor(labels)

class AlignPairCollate(object):

    def __init__(self, imgH=32, imgW=100, keep_ratio_with_pad=False):
        self.imgH = imgH
        self.imgW = imgW
        self.keep_ratio_with_pad = keep_ratio_with_pad

    def __call__(self, batch):
        batch = filter(lambda x: x is not None, batch)
        images1, images2, labels1, labels2 = zip(*batch)

        if self.keep_ratio_with_pad:  # same concept with 'Rosetta' paper
            resized_max_w = self.imgW
            input_channel = 3 if images1[0].mode == 'RGB' else 1
            transform = NormalizePAD((input_channel, self.imgH, resized_max_w))

            resized_images1 = []
            resized_images2 = []
            cntr=0
            for image1 in images1:
                image2=images2[cntr]
                w, h = image1.size
                ratio = w / float(h)
                if math.ceil(self.imgH * ratio) > self.imgW:
                    resized_w = self.imgW
                else:
                    resized_w = math.ceil(self.imgH * ratio)

                resized_image1 = image1.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_images1.append(transform(resized_image1))
                
                #resizing image2 w.r.t image1 dimensions
                resized_image2 = image2.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_images2.append(transform(resized_image2))
                # resized_image.save('./image_test/%d_test.jpg' % w)
                cntr+=1

            image_tensors1 = torch.cat([t.unsqueeze(0) for t in resized_images1], 0)
            image_tensors2 = torch.cat([t.unsqueeze(0) for t in resized_images2], 0)

        else:
            transform = ResizeNormalize((self.imgW, self.imgH))
            image_tensors1 = [transform(image) for image in images1]
            image_tensors2 = [transform(image) for image in images2]
            image_tensors1 = torch.cat([t.unsqueeze(0) for t in image_tensors1], 0)
            image_tensors2 = torch.cat([t.unsqueeze(0) for t in image_tensors2], 0)

        return image_tensors1, image_tensors2, labels1, labels2

class AlignPHOCCollate(object):

    def __init__(self, imgH=32, imgW=100, keep_ratio_with_pad=False):
        self.imgH = imgH
        self.imgW = imgW
        self.keep_ratio_with_pad = keep_ratio_with_pad

    def __call__(self, batch):
        batch = filter(lambda x: x is not None, batch)
        images1, images2, labels1, labels2, phoc1, phoc2 = zip(*batch)

        if self.keep_ratio_with_pad:  # same concept with 'Rosetta' paper
            resized_max_w = self.imgW
            input_channel = 3 if images1[0].mode == 'RGB' else 1
            transform = NormalizePAD((input_channel, self.imgH, resized_max_w))

            resized_images1 = []
            resized_images2 = []
            cntr=0
            for image1 in images1:
                image2=images2[cntr]
                w, h = image1.size
                ratio = w / float(h)
                if math.ceil(self.imgH * ratio) > self.imgW:
                    resized_w = self.imgW
                else:
                    resized_w = math.ceil(self.imgH * ratio)

                resized_image1 = image1.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_images1.append(transform(resized_image1))
                
                #resizing image2 w.r.t image1 dimensions
                resized_image2 = image2.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_images2.append(transform(resized_image2))
                # resized_image.save('./image_test/%d_test.jpg' % w)
                cntr+=1

            image_tensors1 = torch.cat([t.unsqueeze(0) for t in resized_images1], 0)
            image_tensors2 = torch.cat([t.unsqueeze(0) for t in resized_images2], 0)

        else:
            transform = ResizeNormalize((self.imgW, self.imgH))
            image_tensors1 = [transform(image) for image in images1]
            image_tensors2 = [transform(image) for image in images2]
            image_tensors1 = torch.cat([t.unsqueeze(0) for t in image_tensors1], 0)
            image_tensors2 = torch.cat([t.unsqueeze(0) for t in image_tensors2], 0)
        
        phoc1_tensors = torch.cat([torch.tensor(t).unsqueeze(0) for t in phoc1], 0)
        phoc2_tensors = torch.cat([torch.tensor(t).unsqueeze(0) for t in phoc2], 0)

        return image_tensors1, image_tensors2, labels1, labels2, phoc1_tensors, phoc2_tensors

class AlignPairImgCollate(object):

    def __init__(self, imgH=32, imgW=100, keep_ratio_with_pad=False):
        self.imgH = imgH
        self.imgW = imgW
        self.keep_ratio_with_pad = keep_ratio_with_pad

    def __call__(self, batch):
        batch = filter(lambda x: x is not None, batch)
        images1, images2, labels1, labels2, label1SynthImgs, label2SynthImgs = zip(*batch)

        if self.keep_ratio_with_pad:  # same concept with 'Rosetta' paper
            resized_max_w = self.imgW
            input_channel = 3 if images1[0].mode == 'RGB' else 1
            transform = NormalizePAD((input_channel, self.imgH, resized_max_w))

            input_channel_synth = 3 if label1SynthImgs[0].mode == 'RGB' else 1
            transform_synth = NormalizePAD((input_channel_synth, self.imgH, resized_max_w))

            resized_images1 = []
            resized_images2 = []
            resized_label1SynthImgs = []
            resized_label2SynthImgs = []
            
            cntr=0
            for image1 in images1:
                image2=images2[cntr]
                label1SynthImg=label1SynthImgs[cntr]
                label2SynthImg=label2SynthImgs[cntr]
                

                w, h = image1.size
                ratio = w / float(h)
                if math.ceil(self.imgH * ratio) > self.imgW:
                    resized_w = self.imgW
                else:
                    resized_w = math.ceil(self.imgH * ratio)

                resized_image1 = image1.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_images1.append(transform(resized_image1))
                
                #resizing image2 w.r.t image1 dimensions
                resized_image2 = image2.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_images2.append(transform(resized_image2))

                resized_label1SynthImg = label1SynthImg.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_label1SynthImgs.append(transform_synth(resized_label1SynthImg))

                resized_label2SynthImg = label2SynthImg.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_label2SynthImgs.append(transform_synth(resized_label2SynthImg))
                # resized_image.save('./image_test/%d_test.jpg' % w)
                cntr+=1

            image_tensors1 = torch.cat([t.unsqueeze(0) for t in resized_images1], 0)
            image_tensors2 = torch.cat([t.unsqueeze(0) for t in resized_images2], 0)
            label1SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in resized_label1SynthImgs], 0)
            label2SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in resized_label2SynthImgs], 0)

        else:
            transform = ResizeNormalize((self.imgW, self.imgH))
            image_tensors1 = [transform(image) for image in images1]
            image_tensors2 = [transform(image) for image in images2]
            
            label1SynthImg_tensors = [transform(image) for image in label1SynthImgs]
            label2SynthImg_tensors = [transform(image) for image in label2SynthImgs]
            image_tensors1 = torch.cat([t.unsqueeze(0) for t in image_tensors1], 0)
            image_tensors2 = torch.cat([t.unsqueeze(0) for t in image_tensors2], 0)
            label1SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in label1SynthImg_tensors], 0)
            label2SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in label2SynthImg_tensors], 0)
        # print(image_tensors1.shape)
        # print(label1SynthImg_tensors.shape)
        return image_tensors1, image_tensors2, labels1, labels2, label1SynthImg_tensors, label2SynthImg_tensors


class AlignPairImgCollate_dataaug(object):

    def __init__(self, imgH=32, imgW=100, keep_ratio_with_pad=False, aug=False):
        self.imgH = imgH
        self.imgW = imgW
        self.aug = aug
        self.keep_ratio_with_pad = keep_ratio_with_pad

    def __call__(self, batch):
        batch = filter(lambda x: x is not None, batch)
        images1, images2, labels1, labels2, label1SynthImgs, label2SynthImgs = zip(*batch)

        if self.keep_ratio_with_pad:  # same concept with 'Rosetta' paper
            resized_max_w = self.imgW
            input_channel = 3 if images1[0].mode == 'RGB' else 1
            transform = NormalizePAD((input_channel, self.imgH, resized_max_w))

            input_channel_synth = 3 if label1SynthImgs[0].mode == 'RGB' else 1
            transform_synth = NormalizePAD((input_channel_synth, self.imgH, resized_max_w))

            resized_images1 = []
            resized_images2 = []
            resized_label1SynthImgs = []
            resized_label2SynthImgs = []
            
            cntr=0
            for image1 in images1:
                image2=images2[cntr]
                label1SynthImg=label1SynthImgs[cntr]
                label2SynthImg=label2SynthImgs[cntr]
                

                w, h = image1.size
                ratio = w / float(h)
                if math.ceil(self.imgH * ratio) > self.imgW:
                    resized_w = self.imgW
                else:
                    resized_w = math.ceil(self.imgH * ratio)

                resized_image1 = image1.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_images1.append(transform(resized_image1))
                
                #resizing image2 w.r.t image1 dimensions
                resized_image2 = image2.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_images2.append(transform(resized_image2))

                resized_label1SynthImg = label1SynthImg.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_label1SynthImgs.append(transform_synth(resized_label1SynthImg))

                resized_label2SynthImg = label2SynthImg.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_label2SynthImgs.append(transform_synth(resized_label2SynthImg))
                # resized_image.save('./image_test/%d_test.jpg' % w)
                cntr+=1

            image_tensors1 = torch.cat([t.unsqueeze(0) for t in resized_images1], 0)
            image_tensors2 = torch.cat([t.unsqueeze(0) for t in resized_images2], 0)
            label1SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in resized_label1SynthImgs], 0)
            label2SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in resized_label2SynthImgs], 0)

        else:
            transform_source = ResizeNormalize_translate((self.imgW, self.imgH))
            transform = ResizeNormalize((self.imgW, self.imgH))
            if self.aug:
                image_tensors1 = [transform_source(image) for image in images1]
                image_tensors2 = [transform(image) for image in images2]
                
                label1SynthImg_tensors = [transform(image) for image in label1SynthImgs]
                label2SynthImg_tensors = [transform(image) for image in label2SynthImgs]
            else:
                image_tensors1 = [transform(image) for image in images1]
                image_tensors2 = [transform(image) for image in images2]
                
                label1SynthImg_tensors = [transform(image) for image in label1SynthImgs]
                label2SynthImg_tensors = [transform(image) for image in label2SynthImgs]
            
            image_tensors1 = torch.cat([t.unsqueeze(0) for t in image_tensors1], 0)
            image_tensors2 = torch.cat([t.unsqueeze(0) for t in image_tensors2], 0)
            label1SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in label1SynthImg_tensors], 0)
            label2SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in label2SynthImg_tensors], 0)
        # print(image_tensors1.shape)
        # print(label1SynthImg_tensors.shape)
        return image_tensors1, image_tensors2, labels1, labels2, label1SynthImg_tensors, label2SynthImg_tensors




class AlignPairImgCollate_Norm(object):

    def __init__(self, imgH=32, imgW=100, keep_ratio_with_pad=False):
        self.imgH = imgH
        self.imgW = imgW
        self.keep_ratio_with_pad = keep_ratio_with_pad

    def __call__(self, batch):
        batch = filter(lambda x: x is not None, batch)
        images1, images2, labels1, labels2, label1SynthImgs, label2SynthImgs = zip(*batch)

        if self.keep_ratio_with_pad:  # same concept with 'Rosetta' paper
            resized_max_w = self.imgW
            input_channel = 3 if images1[0].mode == 'RGB' else 1
            transform = NormalizePAD_Mean((input_channel, self.imgH, resized_max_w))

            input_channel_synth = 3 if label1SynthImgs[0].mode == 'RGB' else 1
            transform_synth = NormalizePAD((input_channel_synth, self.imgH, resized_max_w))

            resized_images1 = []
            resized_images2 = []
            resized_label1SynthImgs = []
            resized_label2SynthImgs = []
            
            cntr=0
            for image1 in images1:
                image2=images2[cntr]
                label1SynthImg=label1SynthImgs[cntr]
                label2SynthImg=label2SynthImgs[cntr]
                

                w, h = image1.size
                ratio = w / float(h)
                if math.ceil(self.imgH * ratio) > self.imgW:
                    resized_w = self.imgW
                else:
                    resized_w = math.ceil(self.imgH * ratio)

                resized_image1 = image1.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_images1.append(transform(resized_image1))
                
                #resizing image2 w.r.t image1 dimensions
                resized_image2 = image2.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_images2.append(transform(resized_image2))

                resized_label1SynthImg = label1SynthImg.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_label1SynthImgs.append(transform_synth(resized_label1SynthImg))

                resized_label2SynthImg = label2SynthImg.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_label2SynthImgs.append(transform_synth(resized_label2SynthImg))
                # resized_image.save('./image_test/%d_test.jpg' % w)
                cntr+=1

            image_tensors1 = torch.cat([t.unsqueeze(0) for t in resized_images1], 0)
            image_tensors2 = torch.cat([t.unsqueeze(0) for t in resized_images2], 0)
            label1SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in resized_label1SynthImgs], 0)
            label2SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in resized_label2SynthImgs], 0)

        else:
            
            transform = ResizeNormalize_Mean((self.imgW, self.imgH))
            transform_synth = ResizeNormalize((self.imgW, self.imgH))
            
            image_tensors1 = [transform(image) for image in images1]
            image_tensors2 = [transform(image) for image in images2]
            
            label1SynthImg_tensors = [transform_synth(image) for image in label1SynthImgs]
            label2SynthImg_tensors = [transform_synth(image) for image in label2SynthImgs]
            image_tensors1 = torch.cat([t.unsqueeze(0) for t in image_tensors1], 0)
            image_tensors2 = torch.cat([t.unsqueeze(0) for t in image_tensors2], 0)
            label1SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in label1SynthImg_tensors], 0)
            label2SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in label2SynthImg_tensors], 0)
        # print(image_tensors1.shape)
        # print(label1SynthImg_tensors.shape)
        return image_tensors1, image_tensors2, labels1, labels2, label1SynthImg_tensors, label2SynthImg_tensors

class AlignPairImgCollateVarSize(object):

    def __init__(self, imgH=32, imgW=100, keep_ratio_with_pad=False):
        self.imgH = imgH
        self.imgW = imgW
        self.keep_ratio_with_pad = keep_ratio_with_pad

    def __call__(self, batch):
        batch = filter(lambda x: x is not None, batch)
        images1, images2, labels1, labels2, label1SynthImgs, label2SynthImgs = zip(*batch)

        if self.keep_ratio_with_pad:  # same concept with 'Rosetta' paper
            resized_max_w = self.imgW
            input_channel = 3 if images1[0].mode == 'RGB' else 1
            transform = NormalizePADVarSize((input_channel, self.imgH, resized_max_w))

            input_channel_synth = 3 if label1SynthImgs[0].mode == 'RGB' else 1
            transform_synth = NormalizePADVarSize((input_channel_synth, self.imgH, resized_max_w))

            resized_images1 = []
            resized_images2 = []
            resized_label1SynthImgs = []
            resized_label2SynthImgs = []
            
            cntr=0
            for image1 in images1:
                image2=images2[cntr]
                label1SynthImg=label1SynthImgs[cntr]
                label2SynthImg=label2SynthImgs[cntr]
                

                w, h = image1.size
                ratio = w / float(h)
                if math.ceil(self.imgH * ratio) > self.imgW:
                    resized_w = self.imgW
                else:
                    resized_w = math.ceil(self.imgH * ratio)

                resized_image1 = image1.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_images1.append(transform(resized_image1))
                
                #resizing image2 w.r.t image1 dimensions
                resized_image2 = image2.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_images2.append(transform(resized_image2))

                resized_label1SynthImg = label1SynthImg.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_label1SynthImgs.append(transform_synth(resized_label1SynthImg))

                resized_label2SynthImg = label2SynthImg.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_label2SynthImgs.append(transform_synth(resized_label2SynthImg))
                # resized_image.save('./image_test/%d_test.jpg' % w)
                cntr+=1

            image_tensors1 = torch.cat([t.unsqueeze(0) for t in resized_images1], 0)
            image_tensors2 = torch.cat([t.unsqueeze(0) for t in resized_images2], 0)
            label1SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in resized_label1SynthImgs], 0)
            label2SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in resized_label2SynthImgs], 0)

        else:
            transform = ResizeNormalize((self.imgW, self.imgH))
            image_tensors1 = [transform(image) for image in images1]
            image_tensors2 = [transform(image) for image in images2]
            
            label1SynthImg_tensors = [transform(image) for image in label1SynthImgs]
            label2SynthImg_tensors = [transform(image) for image in label2SynthImgs]
            image_tensors1 = torch.cat([t.unsqueeze(0) for t in image_tensors1], 0)
            image_tensors2 = torch.cat([t.unsqueeze(0) for t in image_tensors2], 0)
            label1SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in label1SynthImg_tensors], 0)
            label2SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in label2SynthImg_tensors], 0)
        # print(image_tensors1.shape)
        # print(label1SynthImg_tensors.shape)
        return image_tensors1, image_tensors2, labels1, labels2, label1SynthImg_tensors, label2SynthImg_tensors


class AlignPairImgCollate_Test(object):

    def __init__(self, imgH=32, imgW=100, keep_ratio_with_pad=False):
        self.imgH = imgH
        self.imgW = imgW
        self.keep_ratio_with_pad = keep_ratio_with_pad

    def __call__(self, batch):
        batch = filter(lambda x: x is not None, batch)
        images1, images2, labels1, labels2, label1SynthImgs, label2SynthImgs = zip(*batch)
        images1_shape = []
        images2_shape = []
        
        for image in images1:
            images1_shape.append(image.size)
        for image in images2:
            images2_shape.append(image.size)

        if self.keep_ratio_with_pad:  # same concept with 'Rosetta' paper
            resized_max_w = self.imgW
            input_channel = 3 if images1[0].mode == 'RGB' else 1
            transform = NormalizePAD((input_channel, self.imgH, resized_max_w))

            input_channel_synth = 3 if label1SynthImgs[0].mode == 'RGB' else 1
            transform_synth = NormalizePAD((input_channel_synth, self.imgH, resized_max_w))

            resized_images1 = []
            resized_images2 = []
            resized_label1SynthImgs = []
            resized_label2SynthImgs = []
            
            cntr=0
            for image1 in images1:
                image2=images2[cntr]
                label1SynthImg=label1SynthImgs[cntr]
                label2SynthImg=label2SynthImgs[cntr]
                

                w, h = image1.size
                ratio = w / float(h)
                if math.ceil(self.imgH * ratio) > self.imgW:
                    resized_w = self.imgW
                else:
                    resized_w = math.ceil(self.imgH * ratio)

                resized_image1 = image1.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_images1.append(transform(resized_image1))
                
                #resizing image2 w.r.t image1 dimensions
                resized_image2 = image2.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_images2.append(transform(resized_image2))

                resized_label1SynthImg = label1SynthImg.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_label1SynthImgs.append(transform_synth(resized_label1SynthImg))

                resized_label2SynthImg = label2SynthImg.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_label2SynthImgs.append(transform_synth(resized_label2SynthImg))
                # resized_image.save('./image_test/%d_test.jpg' % w)
                cntr+=1

            image_tensors1 = torch.cat([t.unsqueeze(0) for t in resized_images1], 0)
            image_tensors2 = torch.cat([t.unsqueeze(0) for t in resized_images2], 0)
            label1SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in resized_label1SynthImgs], 0)
            label2SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in resized_label2SynthImgs], 0)

        else:
            transform = ResizeNormalize((self.imgW, self.imgH))
            image_tensors1 = [transform(image) for image in images1]
            image_tensors2 = [transform(image) for image in images2]

            
            label1SynthImg_tensors = [transform(image) for image in label1SynthImgs]
            label2SynthImg_tensors = [transform(image) for image in label2SynthImgs]
            image_tensors1 = torch.cat([t.unsqueeze(0) for t in image_tensors1], 0)
            image_tensors2 = torch.cat([t.unsqueeze(0) for t in image_tensors2], 0)
            label1SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in label1SynthImg_tensors], 0)
            label2SynthImg_tensors = torch.cat([t.unsqueeze(0) for t in label2SynthImg_tensors], 0)
        # print(image_tensors1.shape)
        # print(label1SynthImg_tensors.shape)
        return image_tensors1, image_tensors2, labels1, labels2, label1SynthImg_tensors, label2SynthImg_tensors, images1_shape, images2_shape




def tensor2im(image_tensor, imtype=np.uint8):
    image_numpy = image_tensor.cpu().float().numpy()
    if image_numpy.shape[0] == 1:
        image_numpy = np.tile(image_numpy, (3, 1, 1))
    image_numpy = (np.transpose(image_numpy, (1, 2, 0)) + 1) / 2.0 * 255.0
    return image_numpy.astype(imtype)



def save_image(image_numpy, image_path):
    image_pil = Image.fromarray(image_numpy)
    image_pil.save(image_path)

# PHOC Dataset iterator
class phoc_gen(Dataset):
    def __init__(self, opt):
        
        unigram_levels = [1,2,3,4]
        bigram_levels = [2,3]

        self.word2Idx = {}
        self.idx2Word = {}
        self.words=[]
        strings=[]
        unqStrings=[]

        if os.path.exists(opt.words_file):
            lexFID = open(opt.words_file,'r')
            lines = lexFID.readlines()
            lexFID.close()
        else:
            sys.exit('Lexicon file not found')

        cntr=0
        gCntr=0
        for currWord in lines:
            if opt.sensitive:
                currWord = currWord[:-1]
            else:
                currWord = currWord[:-1].lower()
            
            if not(currWord in self.word2Idx):
                self.word2Idx[currWord] = cntr
                self.idx2Word[cntr] = currWord
                unqStrings.append(currWord)
                cntr+=1
            
            self.words.append(self.word2Idx[currWord])
            strings.append(currWord)
            gCntr+=1
        
        phoc_unigrams = opt.classes
        phoc_bigrams = phoc.get_most_common_n_grams(strings, num_results=50, n=2)
        
        self.phoc_size = len(phoc_unigrams) * np.sum(unigram_levels)
        
        if phoc_bigrams is not None:
            self.phoc_size += len(phoc_bigrams)*np.sum(bigram_levels)

        print('Building PHOC matrix--started')
        self.phocMat = np.zeros((len(self.words),self.phoc_size), dtype=np.float32)
        self.phocMat = phoc.build_phoc(unqStrings, phoc_unigrams, unigram_levels, 
            bigram_levels=bigram_levels, phoc_bigrams=phoc_bigrams, on_unknown_unigram='warn')
        print('Building PHOC matrix--ended')
    
    def __len__(self):
        return len(self.words)

    def __getitem__(self, index):
        return (self.phocMat[self.words[index],:], self.idx2Word[self.words[index]])
    
    def getPhoc(self, word):
        if word in self.word2Idx:
            return self.phocMat[self.word2Idx[word],:]
        else:
            print('warning; phoc vector not found')
            return np.zeros((self.phoc_size),dtype=np.float32)


# PHOC Dataset iterator
class text_gen(Dataset):
    def __init__(self, opt):
        
        

        self.word2Idx = {}
        self.idx2Word = {}
        self.words=[]
        strings=[]
        unqStrings=[]

        if os.path.exists(opt.words_file):
            lexFID = open(opt.words_file,'r')
            lines = lexFID.readlines()
            lexFID.close()
        else:
            sys.exit('Lexicon file not found')

        cntr=0
        gCntr=0
        
        # We only train and evaluate on alphanumerics (or pre-defined character set in train.py)
        out_of_char = f'[^{opt.character}]'
        for currWord in lines:
            if opt.sensitive:
                currWord = currWord[:-1]
            else:
                currWord = currWord[:-1].lower()
            
            currWord = re.sub(out_of_char, '', currWord)
            
            if not(currWord in self.word2Idx):
                self.word2Idx[currWord] = cntr
                self.idx2Word[cntr] = currWord
                unqStrings.append(currWord)
                cntr+=1
            
            self.words.append(self.word2Idx[currWord])
            strings.append(currWord)
            gCntr+=1
        
    
    def __len__(self):
        return len(self.words)

    def __getitem__(self, index):
        return self.idx2Word[self.words[index]]

class text_gen_synth(Dataset):
    def __init__(self, opt):

        self.word2Idx = {}
        self.idx2Word = {}
        self.words=[]
        self.synthGen = SynthGenerator('/private/home/pkrishnan/fonts/fonts-10-path.txt',imgSize=(64,256))

        strings=[]
        unqStrings=[]

        if os.path.exists(opt.words_file):
            lexFID = open(opt.words_file,'r')
            lines = lexFID.readlines()
            lexFID.close()
        else:
            sys.exit('Lexicon file not found')

        cntr=0
        gCntr=0
        
        # We only train and evaluate on alphanumerics (or pre-defined character set in train.py)
        out_of_char = f'[^{opt.character}]'
        for currWord in lines:
            if opt.sensitive:
                currWord = currWord[:-1]
            else:
                currWord = currWord[:-1].lower()
            
            currWord = re.sub(out_of_char, '', currWord)
            
            if not(currWord in self.word2Idx):
                self.word2Idx[currWord] = cntr
                self.idx2Word[cntr] = currWord
                unqStrings.append(currWord)
                cntr+=1
            
            self.words.append(self.word2Idx[currWord])
            strings.append(currWord)
            gCntr+=1
        
    
    def __len__(self):
        return len(self.words)

    def __getitem__(self, index):
        label = self.idx2Word[self.words[index]]  

        return label, self.synthGen.synthesizeWordImage(label, 0)


class AlignSynthTextCollate(object):

    def __init__(self, imgH=32, imgW=100, keep_ratio_with_pad=False):
        self.imgH = imgH
        self.imgW = imgW
        self.keep_ratio_with_pad = keep_ratio_with_pad

    def __call__(self, batch):
        batch = filter(lambda x: x is not None, batch)
        labels, labelSynthImgs = zip(*batch)

        if self.keep_ratio_with_pad:  # same concept with 'Rosetta' paper
            resized_max_w = self.imgW
            input_channel = 3 if labelSynthImgs[0].mode == 'RGB' else 1
            transform = NormalizePAD((input_channel, self.imgH, resized_max_w))

            resized_labelSynthImgs = []

            cntr=0
            for labelSynthImg in labelSynthImgs:
                

                w, h = labelSynthImg.size
                ratio = w / float(h)
                if math.ceil(self.imgH * ratio) > self.imgW:
                    resized_w = self.imgW
                else:
                    resized_w = math.ceil(self.imgH * ratio)

                resized_labelSynthImg = labelSynthImg.resize((resized_w, self.imgH), Image.BICUBIC)
                resized_labelSynthImgs.append(transform(resized_labelSynthImg))
                
                cntr+=1

            labelSynthImg_tensors = torch.cat([t.unsqueeze(0) for t in resized_labelSynthImgs], 0)

        else:
            transform = ResizeNormalize((self.imgW, self.imgH))
            labelSynthImg_tensors = [transform(image) for image in labelSynthImgs]
            labelSynthImg_tensors = torch.cat([t.unsqueeze(0) for t in labelSynthImg_tensors], 0)

        return labels, labelSynthImg_tensors


class fontDataset(Dataset):
    def __init__(self, imgDir, annFile, transform=None, F2Idx=None, Idx2F=None, numClasses=-1):
        self.imgDir = imgDir
        self.imgPath = []
        self.fontIdx = []
        # self.initIdx = False
        self.transform = transform

        if F2Idx:
            self.F2Idx=F2Idx
            self.Idx2F=Idx2F
            # self.initIdx = True
        else:   
            self.F2Idx={}
            self.Idx2F=[]
        
        cntr=0
        fid = open(annFile,'r')

        for line in fid:

            currLine = line[:-1].split()
            

            if F2Idx and not(currLine[4] in F2Idx):
                print('WARNING: font not found...............', line)
                continue
            
            if currLine[4] in self.F2Idx:
                #IMP: to remove/comment

                if numClasses==-1 or self.F2Idx[currLine[4]] < numClasses:
                    self.fontIdx.append(self.F2Idx[currLine[4]])
                    self.imgPath.append(currLine[0])
            else:
                
                self.F2Idx[currLine[4]] = cntr
                self.Idx2F.append(currLine[4])
                #IMP: to remove/comment
                if numClasses==-1 or self.F2Idx[currLine[4]] < numClasses:
                    self.fontIdx.append(cntr)
                    self.imgPath.append(currLine[0])

                cntr+=1

        fid.close()
        
    def __len__(self):
        return len(self.imgPath)

    
    def __getitem__(self, index):

        try:
            img = Image.open(os.path.join(self.imgDir,self.imgPath[index])).convert('RGB')  # for color image
            if self.transform:
                img = self.transform(img)
        except IOError:
            print(f'Corrupted image for {index}')
            # make dummy image and dummy label for corrupted image.
            img = Image.new('RGB', (256, 64))
            if self.transform:
                img = self.transform(img)
        
        return img, torch.from_numpy(np.int64([self.fontIdx[index]]))


class fontTextDataset(Dataset):
    def __init__(self, imgDir, annFile, transform=None, F2Idx=None, Idx2F=None, numClasses=-1):
        self.imgDir = imgDir
        self.imgPath = []
        self.fontIdx = []
        # self.initIdx = False
        self.transform = transform

        if F2Idx:
            self.F2Idx=F2Idx
            self.Idx2F=Idx2F
            # self.initIdx = True
        else:   
            self.F2Idx={}
            self.Idx2F=[]
        
        cntr=0
        fid = open(annFile,'r')

        for line in fid:

            currLine = line[:-1].split()
            

            if F2Idx and not(currLine[2] in F2Idx):
                print('WARNING: font not found...............', line)
                continue
            
            if currLine[2] in self.F2Idx:
                #IMP: to remove/comment

                if numClasses==-1 or self.F2Idx[currLine[2]] < numClasses:
                    self.fontIdx.append(self.F2Idx[currLine[2]])
                    self.imgPath.append(currLine[0])
            else:
                
                self.F2Idx[currLine[2]] = cntr
                self.Idx2F.append(currLine[2])
                #IMP: to remove/comment
                if numClasses==-1 or self.F2Idx[currLine[2]] < numClasses:
                    self.fontIdx.append(cntr)
                    self.imgPath.append(currLine[0])

                cntr+=1

        fid.close()
        
    def __len__(self):
        return len(self.imgPath)

    
    def __getitem__(self, index):

        try:
            img = Image.open(os.path.join(self.imgDir,self.imgPath[index])).convert('RGB')  # for color image
            if self.transform:
                img = self.transform(img)
        except IOError:
            print(f'Corrupted image for {index}')
            # make dummy image and dummy label for corrupted image.
            img = Image.new('RGB', (256, 64))
            if self.transform:
                img = self.transform(img)
        
        return img, torch.from_numpy(np.int64([self.fontIdx[index]]))