Question: I need help for training my custom dataset

[kerberosargos]

Hello firstly thank you.

I am using mmocr. I have create "iiit5k" by using mmocr's as following command

python tools/dataset_converters/prepare_dataset.py IIIT5K --task textrecog --overwrite-cfg

I have created cropped image by using textsnake model as following images

My question is "Can I train your model with those type images as a sentences? "

My mmocr iiit5k's textrecog_train.json file as following

{
	"metainfo": {
		"dataset_type": "TextRecogDataset",
		"task_name": "textrecog"
	},
	"data_list": [
		{
			"instances": [
				{
					"text": "Z NO: 2808"
				}
			],
			"img_path": "textrecog_imgs\\train\\00bc6b1efa4849a79bb2d29e1765f93b-ezsam_2.png"
		},
		{
			"instances": [
				{
					"text": "EKÜ NO: 0001"
				}
			],
			"img_path": "textrecog_imgs\\train\\00bc6b1efa4849a79bb2d29e1765f93b-ezsam_1.png"
		}
	]
}

[Mountchicken]

Hi @kerberosargos
Sorry for the late reply. Of course you can use your data for training. You need to prepare the dataset config like this https://github.com/Mountchicken/Union14M/blob/main/mmocr-dev-1.x/configs/textrecog/_base_/datasets/union14m_train.py

[kerberosargos]

Thank you for your reply. But I can not understand "how to create json structure". Below json structure is true?

[Mountchicken]

The json structure you provide is correct.

{
	"metainfo": {
		"dataset_type": "TextRecogDataset",
		"task_name": "textrecog"
	},
	"data_list": [
		{
			"instances": [
				{
					"text": "Z NO: 2808"
				}
			],
			"img_path": "textrecog_imgs\\train\\00bc6b1efa4849a79bb2d29e1765f93b-ezsam_2.png"
		},
		{
			"instances": [
				{
					"text": "EKÜ NO: 0001"
				}
			],
			"img_path": "textrecog_imgs\\train\\00bc6b1efa4849a79bb2d29e1765f93b-ezsam_1.png"
		}
	]
}

[kerberosargos]

Thank you very much for your support.

[kerberosargos]

Hello again. I have created two json files according to iiit5k dataset json files structure as following

textrecoq_train.json

{
    "metainfo": {
        "dataset_type": "TextRecogDataset",
        "task_name": "textrecog"
    },
    "data_list": [
        {
            "instances": [
                {
                    "text": "MF YAB 15005886"
                }
            ],
            "img_path": "textrecog_imgs/00bc6b1efa4849a79bb2d29e1765f93b-ezsam/0.png"
        },
        {
            "instances": [
                {
                    "text": "EKÜ NO:0001"
                }
            ],
            "img_path": "textrecog_imgs/00bc6b1efa4849a79bb2d29e1765f93b-ezsam/1.png"
        },
        
        ................

textrecog_test.json

{
    "metainfo": {
        "dataset_type": "TextRecogDataset",
        "task_name": "textrecog"
    },
    "data_list": [
        {
            "instances": [
                {
                    "text": "Bu alışverişten kazancınız 3.1 TL."
                }
            ],
            "img_path": "textrecog_imgs/00bc6b1efa4849a79bb2d29e1765f93b-ezsam/4.png"
        },
        {
            "instances": [
                {
                    "text": "Güncel Puan:"
                }
            ],
            "img_path": "textrecog_imgs/00bc6b1efa4849a79bb2d29e1765f93b-ezsam/9.png"
        },
        
       .......................

Train Data Len: 1041
Test Data Len: 260

But I can not understand how to train those file by using mmocr for your model?
Can you help me bit more?

Thank you in advance for you support.

[kerberosargos]

Hello again,

I have changed "maerec_s_union14m.py" files code as following

_base_ = [
    '_base_marec_vit_s.py',
    '../_base_/datasets/union14m_train.py',
    '../_base_/datasets/union14m_benchmark.py',
    '../_base_/datasets/cute80.py',
    '../_base_/datasets/iiit5k.py',
    '../_base_/datasets/svt.py',
    '../_base_/datasets/svtp.py',
    '../_base_/datasets/icdar2013.py',
    '../_base_/datasets/icdar2015.py',
    '../_base_/default_runtime.py',
    '../_base_/schedules/schedule_adamw_cos_10e.py',
]

train_list = [
     _base_.iiit5k_textrecog_train
]

val_list = [
    _base_.iiit5k_textrecog_test
]

test_list = [
    _base_.union14m_benchmark_artistic,
    _base_.union14m_benchmark_multi_oriented,
    _base_.union14m_benchmark_contextless,
    _base_.union14m_benchmark_curve,
    _base_.union14m_benchmark_incomplete,
    _base_.union14m_benchmark_incomplete_ori,
    _base_.union14m_benchmark_multi_words,
    _base_.union14m_benchmark_salient,
    _base_.union14m_benchmark_general,
]

default_hooks = dict(logger=dict(type='LoggerHook', interval=50))

auto_scale_lr = dict(base_batch_size=128)

train_dataset = dict(
    type='ConcatDataset', datasets=train_list, pipeline=_base_.train_pipeline)
val_dataset = dict(
    type='ConcatDataset', datasets=val_list, pipeline=_base_.test_pipeline)
test_dataset = dict(
    type='ConcatDataset', datasets=test_list, pipeline=_base_.test_pipeline)

train_dataloader = dict(
    batch_size=128,
    num_workers=12,
    persistent_workers=True,
    pin_memory=True,
    sampler=dict(type='DefaultSampler', shuffle=True),
    dataset=train_dataset)

val_dataloader = dict(
    batch_size=128,
    num_workers=4,
    persistent_workers=True,
    pin_memory=True,
    drop_last=False,
    sampler=dict(type='DefaultSampler', shuffle=False),
    dataset=val_dataset)

test_dataloader = dict(
    batch_size=128,
    num_workers=4,
    persistent_workers=True,
    pin_memory=True,
    drop_last=False,
    sampler=dict(type='DefaultSampler', shuffle=False),
    dataset=test_dataset)

val_evaluator = dict(
    dataset_prefixes=['IIIT5K'])

test_evaluator = dict(dataset_prefixes=[
    'artistic', 'multi-oriented', 'contextless', 'curve', 'incomplete',
    'incomplete-ori', 'multi-words', 'salient', 'general'
])

After I have put my jsons and images files to as folowing folder path

Train command like this

python tools/train.py configs/textrecog/maerec/maerec_s_union14m.py

Command result is

iiit5k_textrecog_test = dict(
    ann_file='textrecog_test.json',
    data_root='data/iiit5k',
    pipeline=None,
    test_mode=True,
    type='OCRDataset')
iiit5k_textrecog_train = dict(
    ann_file='textrecog_train.json',
    data_root='data/iiit5k',
    pipeline=None,
    type='OCRDataset')
launcher = 'none'
load_from = None
log_level = 'INFO'
log_processor = dict(by_epoch=True, type='LogProcessor', window_size=10)
model = dict(
    backbone=dict(
        depth=12,
        embed_dim=384,
        img_size=(
            32,
            128,
        ),
        mlp_ratio=4.0,
        num_heads=6,
        patch_size=(
            4,
            4,
        ),
        pretrained=None,
        qkv_bias=True,
        type='VisionTransformer'),
    data_preprocessor=dict(
        mean=[
            123.675,
            116.28,
            103.53,
        ],
        std=[
            58.395,
            57.12,
            57.375,
        ],
        type='TextRecogDataPreprocessor'),
    decoder=dict(
        d_embedding=384,
        d_inner=1536,
        d_k=48,
        d_model=384,
        d_v=48,
        dictionary=dict(
            dict_file=
            'C:/Projects/DotNet/Python/ocr/bentasocr/libs/mm/configs/textrecog/maerec/../../../dicts/turkish_digits_symbols_space.txt',
            same_start_end=True,
            type='Dictionary',
            with_end=True,
            with_padding=True,
            with_start=True,
            with_unknown=True),
        max_seq_len=48,
        module_loss=dict(
            ignore_first_char=True, reduction='mean', type='CEModuleLoss'),
        n_head=8,
        n_layers=6,
        postprocessor=dict(type='AttentionPostprocessor'),
        type='MAERecDecoder'),
    type='MAERec')
optim_wrapper = dict(
    optimizer=dict(
        betas=(
            0.9,
            0.999,
        ),
        eps=1e-08,
        lr=0.0004,
        type='AdamW',
        weight_decay=0.01),
    type='OptimWrapper')
param_scheduler = [
    dict(
        T_max=10,
        convert_to_iter_based=True,
        eta_min=4e-06,
        type='CosineAnnealingLR'),
]
randomness = dict(seed=None)
resume = False
svt_textrecog_data_root = 'data/svt'
svt_textrecog_test = dict(
    ann_file='textrecog_test.json',
    data_root='data/svt',
    pipeline=None,
    test_mode=True,
    type='OCRDataset')
svt_textrecog_train = dict(
    ann_file='textrecog_train.json',
    data_root='data/svt',
    pipeline=None,
    type='OCRDataset')
svtp_textrecog_data_root = 'data/svtp'
svtp_textrecog_test = dict(
    ann_file='textrecog_test.json',
    data_root='data/svtp',
    pipeline=None,
    test_mode=True,
    type='OCRDataset')
svtp_textrecog_train = dict(
    ann_file='textrecog_train.json',
    data_root='data/svtp',
    pipeline=None,
    type='OCRDataset')
test_cfg = dict(type='TestLoop')
test_dataloader = dict(
    batch_size=128,
    dataset=dict(
        datasets=[
            dict(
                ann_file=
                'data/Union14M-L/Union14M-Benchmarks/artistic/annotation.json',
                data_prefix=dict(
                    img_path='data/Union14M-L/Union14M-Benchmarks/artistic'),
                pipeline=None,
                test_mode=True,
                type='OCRDataset'),
            dict(
                ann_file=
                'data/Union14M-L/Union14M-Benchmarks/multi_oriented/annotation.json',
                data_prefix=dict(
                    img_path=
                    'data/Union14M-L/Union14M-Benchmarks/multi_oriented'),
                pipeline=None,
                test_mode=True,
                type='OCRDataset'),
            dict(
                ann_file=
                'data/Union14M-L/Union14M-Benchmarks/contextless/annotation.json',
                data_prefix=dict(
                    img_path='data/Union14M-L/Union14M-Benchmarks/contextless'
                ),
                pipeline=None,
                test_mode=True,
                type='OCRDataset'),
            dict(
                ann_file=
                'data/Union14M-L/Union14M-Benchmarks/curve/annotation.json',
                data_prefix=dict(
                    img_path='data/Union14M-L/Union14M-Benchmarks/curve'),
                pipeline=None,
                test_mode=True,
                type='OCRDataset'),
            dict(
                ann_file=
                'data/Union14M-L/Union14M-Benchmarks/incomplete/annotation.json',
                data_prefix=dict(
                    img_path='data/Union14M-L/Union14M-Benchmarks/incomplete'),
                pipeline=None,
                test_mode=True,
                type='OCRDataset'),
            dict(
                ann_file=
                'data/Union14M-L/Union14M-Benchmarks/incomplete_ori/annotation.json',
                data_prefix=dict(
                    img_path=
                    'data/Union14M-L/Union14M-Benchmarks/incomplete_ori'),
                pipeline=None,
                test_mode=True,
                type='OCRDataset'),
            dict(
                ann_file=
                'data/Union14M-L/Union14M-Benchmarks/multi_words/annotation.json',
                data_prefix=dict(
                    img_path='data/Union14M-L/Union14M-Benchmarks/multi_words'
                ),
                pipeline=None,
                test_mode=True,
                type='OCRDataset'),
            dict(
                ann_file=
                'data/Union14M-L/Union14M-Benchmarks/salient/annotation.json',
                data_prefix=dict(
                    img_path='data/Union14M-L/Union14M-Benchmarks/salient'),
                pipeline=None,
                test_mode=True,
                type='OCRDataset'),
            dict(
                ann_file=
                'data/Union14M-L/Union14M-Benchmarks/general/annotation.json',
                data_prefix=dict(img_path='data/Union14M-L//'),
                pipeline=None,
                test_mode=True,
                type='OCRDataset'),
        ],
        pipeline=[
            dict(type='LoadImageFromFile'),
            dict(scale=(
                128,
                32,
            ), type='Resize'),
            dict(type='LoadOCRAnnotations', with_text=True),
            dict(
                meta_keys=(
                    'img_path',
                    'ori_shape',
                    'img_shape',
                    'valid_ratio',
                ),
                type='PackTextRecogInputs'),
        ],
        type='ConcatDataset'),
    drop_last=False,
    num_workers=4,
    persistent_workers=True,
    pin_memory=True,
    sampler=dict(shuffle=False, type='DefaultSampler'))
test_dataset = dict(
    datasets=[
        dict(
            ann_file=
            'data/Union14M-L/Union14M-Benchmarks/artistic/annotation.json',
            data_prefix=dict(
                img_path='data/Union14M-L/Union14M-Benchmarks/artistic'),
            pipeline=None,
            test_mode=True,
            type='OCRDataset'),
        dict(
            ann_file=
            'data/Union14M-L/Union14M-Benchmarks/multi_oriented/annotation.json',
            data_prefix=dict(
                img_path='data/Union14M-L/Union14M-Benchmarks/multi_oriented'),
            pipeline=None,
            test_mode=True,
            type='OCRDataset'),
        dict(
            ann_file=
            'data/Union14M-L/Union14M-Benchmarks/contextless/annotation.json',
            data_prefix=dict(
                img_path='data/Union14M-L/Union14M-Benchmarks/contextless'),
            pipeline=None,
            test_mode=True,
            type='OCRDataset'),
        dict(
            ann_file=
            'data/Union14M-L/Union14M-Benchmarks/curve/annotation.json',
            data_prefix=dict(
                img_path='data/Union14M-L/Union14M-Benchmarks/curve'),
            pipeline=None,
            test_mode=True,
            type='OCRDataset'),
        dict(
            ann_file=
            'data/Union14M-L/Union14M-Benchmarks/incomplete/annotation.json',
            data_prefix=dict(
                img_path='data/Union14M-L/Union14M-Benchmarks/incomplete'),
            pipeline=None,
            test_mode=True,
            type='OCRDataset'),
        dict(
            ann_file=
            'data/Union14M-L/Union14M-Benchmarks/incomplete_ori/annotation.json',
            data_prefix=dict(
                img_path='data/Union14M-L/Union14M-Benchmarks/incomplete_ori'),
            pipeline=None,
            test_mode=True,
            type='OCRDataset'),
        dict(
            ann_file=
            'data/Union14M-L/Union14M-Benchmarks/multi_words/annotation.json',
            data_prefix=dict(
                img_path='data/Union14M-L/Union14M-Benchmarks/multi_words'),
            pipeline=None,
            test_mode=True,
            type='OCRDataset'),
        dict(
            ann_file=
            'data/Union14M-L/Union14M-Benchmarks/salient/annotation.json',
            data_prefix=dict(
                img_path='data/Union14M-L/Union14M-Benchmarks/salient'),
            pipeline=None,
            test_mode=True,
            type='OCRDataset'),
        dict(
            ann_file=
            'data/Union14M-L/Union14M-Benchmarks/general/annotation.json',
            data_prefix=dict(img_path='data/Union14M-L//'),
            pipeline=None,
            test_mode=True,
            type='OCRDataset'),
    ],
    pipeline=[
        dict(type='LoadImageFromFile'),
        dict(scale=(
            128,
            32,
        ), type='Resize'),
        dict(type='LoadOCRAnnotations', with_text=True),
        dict(
            meta_keys=(
                'img_path',
                'ori_shape',
                'img_shape',
                'valid_ratio',
            ),
            type='PackTextRecogInputs'),
    ],
    type='ConcatDataset')
test_evaluator = dict(
    dataset_prefixes=[
        'artistic',
        'multi-oriented',
        'contextless',
        'curve',
        'incomplete',
        'incomplete-ori',
        'multi-words',
        'salient',
        'general',
    ],
    metrics=[
        dict(
            mode=[
                'exact',
                'ignore_case',
                'ignore_case_symbol',
            ],
            type='WordMetric'),
        dict(type='CharMetric'),
    ],
    type='MultiDatasetsEvaluator')
test_list = [
    dict(
        ann_file='data/Union14M-L/Union14M-Benchmarks/artistic/annotation.json',
        data_prefix=dict(
            img_path='data/Union14M-L/Union14M-Benchmarks/artistic'),
        pipeline=None,
        test_mode=True,
        type='OCRDataset'),
    dict(
        ann_file=
        'data/Union14M-L/Union14M-Benchmarks/multi_oriented/annotation.json',
        data_prefix=dict(
            img_path='data/Union14M-L/Union14M-Benchmarks/multi_oriented'),
        pipeline=None,
        test_mode=True,
        type='OCRDataset'),
    dict(
        ann_file=
        'data/Union14M-L/Union14M-Benchmarks/contextless/annotation.json',
        data_prefix=dict(
            img_path='data/Union14M-L/Union14M-Benchmarks/contextless'),
        pipeline=None,
        test_mode=True,
        type='OCRDataset'),
    dict(
        ann_file='data/Union14M-L/Union14M-Benchmarks/curve/annotation.json',
        data_prefix=dict(img_path='data/Union14M-L/Union14M-Benchmarks/curve'),
        pipeline=None,
        test_mode=True,
        type='OCRDataset'),
    dict(
        ann_file=
        'data/Union14M-L/Union14M-Benchmarks/incomplete/annotation.json',
        data_prefix=dict(
            img_path='data/Union14M-L/Union14M-Benchmarks/incomplete'),
        pipeline=None,
        test_mode=True,
        type='OCRDataset'),
    dict(
        ann_file=
        'data/Union14M-L/Union14M-Benchmarks/incomplete_ori/annotation.json',
        data_prefix=dict(
            img_path='data/Union14M-L/Union14M-Benchmarks/incomplete_ori'),
        pipeline=None,
        test_mode=True,
        type='OCRDataset'),
    dict(
        ann_file=
        'data/Union14M-L/Union14M-Benchmarks/multi_words/annotation.json',
        data_prefix=dict(
            img_path='data/Union14M-L/Union14M-Benchmarks/multi_words'),
        pipeline=None,
        test_mode=True,
        type='OCRDataset'),
    dict(
        ann_file='data/Union14M-L/Union14M-Benchmarks/salient/annotation.json',
        data_prefix=dict(
            img_path='data/Union14M-L/Union14M-Benchmarks/salient'),
        pipeline=None,
        test_mode=True,
        type='OCRDataset'),
    dict(
        ann_file='data/Union14M-L/Union14M-Benchmarks/general/annotation.json',
        data_prefix=dict(img_path='data/Union14M-L//'),
        pipeline=None,
        test_mode=True,
        type='OCRDataset'),
]
test_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(scale=(
        128,
        32,
    ), type='Resize'),
    dict(type='LoadOCRAnnotations', with_text=True),
    dict(
        meta_keys=(
            'img_path',
            'ori_shape',
            'img_shape',
            'valid_ratio',
        ),
        type='PackTextRecogInputs'),
]
train_cfg = dict(max_epochs=10, type='EpochBasedTrainLoop', val_interval=1)
train_dataloader = dict(
    batch_size=128,
    dataset=dict(
        datasets=[
            dict(
                ann_file='textrecog_train.json',
                data_root='data/iiit5k',
                pipeline=None,
                type='OCRDataset'),
        ],
        pipeline=[
            dict(ignore_empty=True, min_size=0, type='LoadImageFromFile'),
            dict(type='LoadOCRAnnotations', with_text=True),
            dict(scale=(
                128,
                32,
            ), type='Resize'),
            dict(
                prob=0.5,
                transforms=[
                    dict(
                        transforms=[
                            dict(max_angle=15, type='RandomRotate'),
                            dict(
                                degrees=15,
                                op='RandomAffine',
                                scale=(
                                    0.5,
                                    2.0,
                                ),
                                shear=(
                                    -45,
                                    45,
                                ),
                                translate=(
                                    0.3,
                                    0.3,
                                ),
                                type='TorchVisionWrapper'),
                            dict(
                                distortion_scale=0.5,
                                op='RandomPerspective',
                                p=1,
                                type='TorchVisionWrapper'),
                        ],
                        type='RandomChoice'),
                ],
                type='RandomApply'),
            dict(
                prob=0.25,
                transforms=[
                    dict(type='PyramidRescale'),
                    dict(
                        transforms=[
                            dict(
                                p=0.5, type='GaussNoise', var_limit=(
                                    20,
                                    20,
                                )),
                            dict(blur_limit=7, p=0.5, type='MotionBlur'),
                        ],
                        type='mmdet.Albu'),
                ],
                type='RandomApply'),
            dict(
                prob=0.25,
                transforms=[
                    dict(
                        brightness=0.5,
                        contrast=0.5,
                        hue=0.1,
                        op='ColorJitter',
                        saturation=0.5,
                        type='TorchVisionWrapper'),
                ],
                type='RandomApply'),
            dict(
                meta_keys=(
                    'img_path',
                    'ori_shape',
                    'img_shape',
                    'valid_ratio',
                ),
                type='PackTextRecogInputs'),
        ],
        type='ConcatDataset'),
    num_workers=12,
    persistent_workers=True,
    pin_memory=True,
    sampler=dict(shuffle=True, type='DefaultSampler'))
train_dataset = dict(
    datasets=[
        dict(
            ann_file='textrecog_train.json',
            data_root='data/iiit5k',
            pipeline=None,
            type='OCRDataset'),
    ],
    pipeline=[
        dict(ignore_empty=True, min_size=0, type='LoadImageFromFile'),
        dict(type='LoadOCRAnnotations', with_text=True),
        dict(scale=(
            128,
            32,
        ), type='Resize'),
        dict(
            prob=0.5,
            transforms=[
                dict(
                    transforms=[
                        dict(max_angle=15, type='RandomRotate'),
                        dict(
                            degrees=15,
                            op='RandomAffine',
                            scale=(
                                0.5,
                                2.0,
                            ),
                            shear=(
                                -45,
                                45,
                            ),
                            translate=(
                                0.3,
                                0.3,
                            ),
                            type='TorchVisionWrapper'),
                        dict(
                            distortion_scale=0.5,
                            op='RandomPerspective',
                            p=1,
                            type='TorchVisionWrapper'),
                    ],
                    type='RandomChoice'),
            ],
            type='RandomApply'),
        dict(
            prob=0.25,
            transforms=[
                dict(type='PyramidRescale'),
                dict(
                    transforms=[
                        dict(p=0.5, type='GaussNoise', var_limit=(
                            20,
                            20,
                        )),
                        dict(blur_limit=7, p=0.5, type='MotionBlur'),
                    ],
                    type='mmdet.Albu'),
            ],
            type='RandomApply'),
        dict(
            prob=0.25,
            transforms=[
                dict(
                    brightness=0.5,
                    contrast=0.5,
                    hue=0.1,
                    op='ColorJitter',
                    saturation=0.5,
                    type='TorchVisionWrapper'),
            ],
            type='RandomApply'),
        dict(
            meta_keys=(
                'img_path',
                'ori_shape',
                'img_shape',
                'valid_ratio',
            ),
            type='PackTextRecogInputs'),
    ],
    type='ConcatDataset')
train_list = [
    dict(
        ann_file='textrecog_train.json',
        data_root='data/iiit5k',
        pipeline=None,
        type='OCRDataset'),
]
train_pipeline = [
    dict(ignore_empty=True, min_size=0, type='LoadImageFromFile'),
    dict(type='LoadOCRAnnotations', with_text=True),
    dict(scale=(
        128,
        32,
    ), type='Resize'),
    dict(
        prob=0.5,
        transforms=[
            dict(
                transforms=[
                    dict(max_angle=15, type='RandomRotate'),
                    dict(
                        degrees=15,
                        op='RandomAffine',
                        scale=(
                            0.5,
                            2.0,
                        ),
                        shear=(
                            -45,
                            45,
                        ),
                        translate=(
                            0.3,
                            0.3,
                        ),
                        type='TorchVisionWrapper'),
                    dict(
                        distortion_scale=0.5,
                        op='RandomPerspective',
                        p=1,
                        type='TorchVisionWrapper'),
                ],
                type='RandomChoice'),
        ],
        type='RandomApply'),
    dict(
        prob=0.25,
        transforms=[
            dict(type='PyramidRescale'),
            dict(
                transforms=[
                    dict(p=0.5, type='GaussNoise', var_limit=(
                        20,
                        20,
                    )),
                    dict(blur_limit=7, p=0.5, type='MotionBlur'),
                ],
                type='mmdet.Albu'),
        ],
        type='RandomApply'),
    dict(
        prob=0.25,
        transforms=[
            dict(
                brightness=0.5,
                contrast=0.5,
                hue=0.1,
                op='ColorJitter',
                saturation=0.5,
                type='TorchVisionWrapper'),
        ],
        type='RandomApply'),
    dict(
        meta_keys=(
            'img_path',
            'ori_shape',
            'img_shape',
            'valid_ratio',
        ),
        type='PackTextRecogInputs'),
]
tta_model = dict(type='EncoderDecoderRecognizerTTAModel')
tta_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(
        transforms=[
            [
                dict(
                    condition="results['img_shape'][1]<results['img_shape'][0]",
                    true_transforms=[
                        dict(
                            args=[
                                dict(cls='Rot90', k=0, keep_size=False),
                            ],
                            type='ImgAugWrapper'),
                    ],
                    type='ConditionApply'),
                dict(
                    condition="results['img_shape'][1]<results['img_shape'][0]",
                    true_transforms=[
                        dict(
                            args=[
                                dict(cls='Rot90', k=1, keep_size=False),
                            ],
                            type='ImgAugWrapper'),
                    ],
                    type='ConditionApply'),
                dict(
                    condition="results['img_shape'][1]<results['img_shape'][0]",
                    true_transforms=[
                        dict(
                            args=[
                                dict(cls='Rot90', k=3, keep_size=False),
                            ],
                            type='ImgAugWrapper'),
                    ],
                    type='ConditionApply'),
            ],
            [
                dict(scale=(
                    128,
                    32,
                ), type='Resize'),
            ],
            [
                dict(type='LoadOCRAnnotations', with_text=True),
            ],
            [
                dict(
                    meta_keys=(
                        'img_path',
                        'ori_shape',
                        'img_shape',
                        'valid_ratio',
                    ),
                    type='PackTextRecogInputs'),
            ],
        ],
        type='TestTimeAug'),
]
union14m_benchmark_artistic = dict(
    ann_file='data/Union14M-L/Union14M-Benchmarks/artistic/annotation.json',
    data_prefix=dict(img_path='data/Union14M-L/Union14M-Benchmarks/artistic'),
    pipeline=None,
    test_mode=True,
    type='OCRDataset')
union14m_benchmark_contextless = dict(
    ann_file='data/Union14M-L/Union14M-Benchmarks/contextless/annotation.json',
    data_prefix=dict(
        img_path='data/Union14M-L/Union14M-Benchmarks/contextless'),
    pipeline=None,
    test_mode=True,
    type='OCRDataset')
union14m_benchmark_curve = dict(
    ann_file='data/Union14M-L/Union14M-Benchmarks/curve/annotation.json',
    data_prefix=dict(img_path='data/Union14M-L/Union14M-Benchmarks/curve'),
    pipeline=None,
    test_mode=True,
    type='OCRDataset')
union14m_benchmark_general = dict(
    ann_file='data/Union14M-L/Union14M-Benchmarks/general/annotation.json',
    data_prefix=dict(img_path='data/Union14M-L//'),
    pipeline=None,
    test_mode=True,
    type='OCRDataset')
union14m_benchmark_incomplete = dict(
    ann_file='data/Union14M-L/Union14M-Benchmarks/incomplete/annotation.json',
    data_prefix=dict(
        img_path='data/Union14M-L/Union14M-Benchmarks/incomplete'),
    pipeline=None,
    test_mode=True,
    type='OCRDataset')
union14m_benchmark_incomplete_ori = dict(
    ann_file=
    'data/Union14M-L/Union14M-Benchmarks/incomplete_ori/annotation.json',
    data_prefix=dict(
        img_path='data/Union14M-L/Union14M-Benchmarks/incomplete_ori'),
    pipeline=None,
    test_mode=True,
    type='OCRDataset')
union14m_benchmark_multi_oriented = dict(
    ann_file=
    'data/Union14M-L/Union14M-Benchmarks/multi_oriented/annotation.json',
    data_prefix=dict(
        img_path='data/Union14M-L/Union14M-Benchmarks/multi_oriented'),
    pipeline=None,
    test_mode=True,
    type='OCRDataset')
union14m_benchmark_multi_words = dict(
    ann_file='data/Union14M-L/Union14M-Benchmarks/multi_words/annotation.json',
    data_prefix=dict(
        img_path='data/Union14M-L/Union14M-Benchmarks/multi_words'),
    pipeline=None,
    test_mode=True,
    type='OCRDataset')
union14m_benchmark_root = 'data/Union14M-L/Union14M-Benchmarks'
union14m_benchmark_salient = dict(
    ann_file='data/Union14M-L/Union14M-Benchmarks/salient/annotation.json',
    data_prefix=dict(img_path='data/Union14M-L/Union14M-Benchmarks/salient'),
    pipeline=None,
    test_mode=True,
    type='OCRDataset')
union14m_challenging = dict(
    ann_file='train_annos/mmocr1.0/train_challenging.json',
    data_root='data/Union14M-L/',
    pipeline=None,
    test_mode=True,
    type='OCRDataset')
union14m_data_root = 'data/Union14M-L/'
union14m_easy = dict(
    ann_file='train_annos/mmocr1.0/train_easy.json',
    data_root='data/Union14M-L/',
    pipeline=None,
    type='OCRDataset')
union14m_hard = dict(
    ann_file='train_annos/mmocr1.0/train_hard.json',
    data_root='data/Union14M-L/',
    pipeline=None,
    type='OCRDataset')
union14m_medium = dict(
    ann_file='train_annos/mmocr1.0/train_medium.json',
    data_root='data/Union14M-L/',
    pipeline=None,
    type='OCRDataset')
union14m_normal = dict(
    ann_file='train_annos/mmocr1.0/train_normal.json',
    data_root='data/Union14M-L/',
    pipeline=None,
    type='OCRDataset')
union14m_root = 'data/Union14M-L/'
union14m_val = dict(
    ann_file='train_annos/mmocr1.0/val_annos.json',
    data_root='data/Union14M-L/',
    pipeline=None,
    type='OCRDataset')
val_cfg = dict(type='ValLoop')
val_dataloader = dict(
    batch_size=128,
    dataset=dict(
        datasets=[
            dict(
                ann_file='textrecog_test.json',
                data_root='data/iiit5k',
                pipeline=None,
                test_mode=True,
                type='OCRDataset'),
        ],
        pipeline=[
            dict(type='LoadImageFromFile'),
            dict(scale=(
                128,
                32,
            ), type='Resize'),
            dict(type='LoadOCRAnnotations', with_text=True),
            dict(
                meta_keys=(
                    'img_path',
                    'ori_shape',
                    'img_shape',
                    'valid_ratio',
                ),
                type='PackTextRecogInputs'),
        ],
        type='ConcatDataset'),
    drop_last=False,
    num_workers=4,
    persistent_workers=True,
    pin_memory=True,
    sampler=dict(shuffle=False, type='DefaultSampler'))
val_dataset = dict(
    datasets=[
        dict(
            ann_file='textrecog_test.json',
            data_root='data/iiit5k',
            pipeline=None,
            test_mode=True,
            type='OCRDataset'),
    ],
    pipeline=[
        dict(type='LoadImageFromFile'),
        dict(scale=(
            128,
            32,
        ), type='Resize'),
        dict(type='LoadOCRAnnotations', with_text=True),
        dict(
            meta_keys=(
                'img_path',
                'ori_shape',
                'img_shape',
                'valid_ratio',
            ),
            type='PackTextRecogInputs'),
    ],
    type='ConcatDataset')
val_evaluator = dict(
    dataset_prefixes=[
        'IIIT5K',
    ],
    metrics=[
        dict(
            mode=[
                'exact',
                'ignore_case',
                'ignore_case_symbol',
            ],
            type='WordMetric'),
        dict(type='CharMetric'),
    ],
    type='MultiDatasetsEvaluator')
val_list = [
    dict(
        ann_file='textrecog_test.json',
        data_root='data/iiit5k',
        pipeline=None,
        test_mode=True,
        type='OCRDataset'),
]
vis_backends = [
    dict(type='LocalVisBackend'),
]
visualizer = dict(
    name='visualizer',
    type='TextRecogLocalVisualizer',
    vis_backends=[
        dict(type='LocalVisBackend'),
    ])
work_dir = './work_dirs\\bentas_maerec_s_union14m'

04/09 12:18:51 - mmengine - INFO - Distributed training is not used, all SyncBatchNorm (SyncBN) layers in the model will be automatically reverted to BatchNormXd layers if they are used.
04/09 12:18:51 - mmengine - INFO - Hooks will be executed in the following order:
before_run:
(VERY_HIGH   ) RuntimeInfoHook
(BELOW_NORMAL) LoggerHook
 --------------------
before_train:
(VERY_HIGH   ) RuntimeInfoHook
(NORMAL      ) IterTimerHook
(VERY_LOW    ) CheckpointHook
 --------------------
before_train_epoch:
(VERY_HIGH   ) RuntimeInfoHook
(NORMAL      ) IterTimerHook
(NORMAL      ) DistSamplerSeedHook
 --------------------
before_train_iter:
(VERY_HIGH   ) RuntimeInfoHook
(NORMAL      ) IterTimerHook
 --------------------
after_train_iter:
(VERY_HIGH   ) RuntimeInfoHook
(NORMAL      ) IterTimerHook
(BELOW_NORMAL) LoggerHook
(LOW         ) ParamSchedulerHook
(VERY_LOW    ) CheckpointHook
 --------------------
after_train_epoch:
(NORMAL      ) IterTimerHook
(NORMAL      ) SyncBuffersHook
(LOW         ) ParamSchedulerHook
(VERY_LOW    ) CheckpointHook
 --------------------
before_val:
(VERY_HIGH   ) RuntimeInfoHook
 --------------------
before_val_epoch:
(NORMAL      ) IterTimerHook
(NORMAL      ) SyncBuffersHook
 --------------------
before_val_iter:
(NORMAL      ) IterTimerHook
 --------------------
after_val_iter:
(NORMAL      ) IterTimerHook
(NORMAL      ) VisualizationHook
(BELOW_NORMAL) LoggerHook
 --------------------
after_val_epoch:
(VERY_HIGH   ) RuntimeInfoHook
(NORMAL      ) IterTimerHook
(BELOW_NORMAL) LoggerHook
(LOW         ) ParamSchedulerHook
(VERY_LOW    ) CheckpointHook
 --------------------
after_val:
(VERY_HIGH   ) RuntimeInfoHook
 --------------------
after_train:
(VERY_HIGH   ) RuntimeInfoHook
(VERY_LOW    ) CheckpointHook
 --------------------
before_test:
(VERY_HIGH   ) RuntimeInfoHook
 --------------------
before_test_epoch:
(NORMAL      ) IterTimerHook
 --------------------
before_test_iter:
(NORMAL      ) IterTimerHook
 --------------------
after_test_iter:
(NORMAL      ) IterTimerHook
(NORMAL      ) VisualizationHook
(BELOW_NORMAL) LoggerHook
 --------------------
after_test_epoch:
(VERY_HIGH   ) RuntimeInfoHook
(NORMAL      ) IterTimerHook
(BELOW_NORMAL) LoggerHook
 --------------------
after_test:
(VERY_HIGH   ) RuntimeInfoHook
 --------------------
after_run:
(BELOW_NORMAL) LoggerHook
 --------------------
04/09 11:50:16 - mmengine - WARNING - "FileClient" will be deprecated in future. Please use io functions in https://mmengine.readthedocs.io/en/latest/api/fileio.html#file-io
04/09 11:50:16 - mmengine - WARNING - "HardDiskBackend" is the alias of "LocalBackend" and the former will be deprecated in future.
04/09 11:50:16 - mmengine - INFO - Checkpoints will be saved to C:\Projects\DotNet\Python\ocr\bentasocr\libs\mm\work_dirs\bentas_maerec_s_union14m. 
04/09 11:54:20 - mmengine - INFO - Exp name: bentas_maerec_s_union14m_20240409_115001
04/09 11:54:20 - mmengine - INFO - Epoch(train)  [1][9/9]  lr: 3.9233e-04  eta: 0:36:34  time: 27.0981  data_time: 7.0543  loss: 3.8230  loss_ce: 3.8230
04/09 11:54:20 - mmengine - INFO - Saving checkpoint at 1 epochs
04/09 11:56:44 - mmengine - INFO - Epoch(val) [1][3/3]    IIIT5K/recog/word_acc: 0.0000  IIIT5K/recog/word_acc_ignore_case: 0.0000  IIIT5K/recog/word_acc_ignore_case_symbol: 0.0000  IIIT5K/recog/char_recall: 0.1536  IIIT5K/recog/char_precision: 0.0373  data_time: 3.4783  time: 46.5298

After train I have tested epoch_1.pth weigth by using mmocr. Result is as following :(

Thank you in advance for your help.

[Mountchicken]

Hi @kerberosargos
Sorry for the late reply, the training process looks fine to me, but the prediction result seems bad. I noticed that you only have 1041 images for training. This is far not enough to train a text recognizer. Typically we need more than 10k images to train a text recognizer.

[kerberosargos]

Hi @kerberosargos Sorry for the late reply, the training process looks fine to me, but the prediction result seems bad. I noticed that you only have 1041 images for training. This is far not enough to train a text recognizer. Typically we need more than 10k images to train a text recognizer.

You are welcome. I thank you for your support.

I would like to ask a question more? 1 epoch is normal proccessing or not?

I will try to train 10K image. Thank you again.

[Mountchicken]

You can train longer. E.g. 30 epochs for 10K image.

[kerberosargos]

Hello again. I would like to ask one more question for understanding training process clearly.

I have detected my own training images by using TextSnake pretrained model.

Images results are as following.

json file of those images is as following

{
    "data_list": [
        {
            "instances": [
                {
                    "text": "alBaraka"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/0.png"
        },
        {
            "instances": [
                {
                    "text": "BU BELGEYİ SAKLAYINIZ"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/1.png"
        },
        {
            "instances": [
                {
                    "text": "BU İŞLEM YURT İÇİ KARTLA YAPILMIŞTIR"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/2.png"
        },
        {
            "instances": [
                {
                    "text": "Uer.: 92.09.09"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/3.png"
        },
        {
            "instances": [
                {
                    "text": "VISA"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/4.png"
        },
        {
            "instances": [
                {
                    "text": "AID:A0000000031010"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/5.png"
        },
        {
            "instances": [
                {
                    "text": "REF NO: 4380536323"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/6.png"
        },
        {
            "instances": [
                {
                    "text": "GRUP NO: 034"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/7.png"
        },
        {
            "instances": [
                {
                    "text": "SN:0013 ONAY KODU: 578529"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/8.png"
        },
        {
            "instances": [
                {
                    "text": "KARŞILIĞI MAL/HİZM ALDIM"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/9.png"
        },
        {
            "instances": [
                {
                    "text": "))))"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/10.png"
        },
        {
            "instances": [
                {
                    "text": "VISA CONTACTLESS"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/11.png"
        },
        {
            "instances": [
                {
                    "text": "İŞLEM TEMASSIZ YAPILMIŞTIR"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/12.png"
        },
        {
            "instances": [
                {
                    "text": "200.00TL"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/13.png"
        },
        {
            "instances": [
                {
                    "text": "TUTAR"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/14.png"
        },
        {
            "instances": [
                {
                    "text": "**** **** **** 8900"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/15.png"
        },
        {
            "instances": [
                {
                    "text": "18:17:51 Q ONLINE"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/16.png"
        },
        {
            "instances": [
                {
                    "text": "SATIŞ"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/17.png"
        },
        {
            "instances": [
                {
                    "text": "MÜŞTERİ NÜSHASI"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/18.png"
        },
        {
            "instances": [
                {
                    "text": "00A433B2"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/19.png"
        },
        {
            "instances": [
                {
                    "text": "TERMİNAL NO"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/20.png"
        },
        {
            "instances": [
                {
                    "text": "2602185003"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/21.png"
        },
        {
            "instances": [
                {
                    "text": "İŞYERİ NO"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/22.png"
        },
        {
            "instances": [
                {
                    "text": "*200,00"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/23.png"
        },
        {
            "instances": [
                {
                    "text": "KREDİ KARTI"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/24.png"
        },
        {
            "instances": [
                {
                    "text": "TOPLAM"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/25.png"
        },
        {
            "instances": [
                {
                    "text": "*200,00"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/26.png"
        },
        {
            "instances": [
                {
                    "text": "TOPKDV"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/27.png"
        },
        {
            "instances": [
                {
                    "text": "*1,98"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/28.png"
        },
        {
            "instances": [
                {
                    "text": "KASAP"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/29.png"
        },
        {
            "instances": [
                {
                    "text": "%1"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/30.png"
        },
        {
            "instances": [
                {
                    "text": "*200,00"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/31.png"
        },
        {
            "instances": [
                {
                    "text": "SAAT: 18:17"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/32.png"
        },
        {
            "instances": [
                {
                    "text": "26-03-2024"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/33.png"
        },
        {
            "instances": [
                {
                    "text": "FİŞ NO: 14"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/34.png"
        },
        {
            "instances": [
                {
                    "text": "TEŞEKKÜRLER"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/35.png"
        },
        {
            "instances": [
                {
                    "text": ""
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/36.png"
        },
        {
            "instances": [
                {
                    "text": "TEPECİK VD: 7850693945"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/37.png"
        },
        {
            "instances": [
                {
                    "text": "NO: 24/B BAŞİSKELE/KOCAELİ"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/38.png"
        },
        {
            "instances": [
                {
                    "text": "SEYMEN MAH D-130 KARAYOLU CAD"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/39.png"
        },
        {
            "instances": [
                {
                    "text": "TURİZM NAK LTD ŞTİ"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/40.png"
        },
        {
            "instances": [
                {
                    "text": "SÜLEYMAN KESKİN RESTORAN İNŞ"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/41.png"
        },
        {
            "instances": [
                {
                    "text": "MODA KEBAP LAHMACUN"
                }
            ],
            "img_path": "7f2ecb4425a847e8b1c20d14e74b7a3f-ezsam/42.png"
        }
    ]
}

So my question is about "images and json file is correct for training?"

Thank you in advance for your support.