text recognition model training problems

[wyy19990]

Tried to retrain with the sar model of mmocr0.x, without changing the model configuration, but the results of one round of training were poor:0_char_recall: 0.2002, 0_char_precision: 0.2004, 0_word_acc: 0.0094, 0_word_acc_ignore_case: 0.0094, 0_word_acc_ignore_case_symbol: 0.0106, 0_1-N.E.D: 0.1162

data config：

dataset_type = 'OCRDataset'

root = 'data/iiit5k'
train_anno_file1 = f'{root}/train_label.txt'

dict_file = 'data/recog/en_dict.txt'

train1 = dict(
type=dataset_type,
img_prefix=f'{root}/imgs/train',
ann_file=train_anno_file1,
loader=dict(
type='AnnFileLoader',
repeat=100,
file_format='txt',
file_storage_backend='disk',
parser=dict(
type='LineStrParser',
keys=['filename', 'text'],
keys_idx=[0, 1],
separator=' ')),
pipeline=None,
test_mode=False)

test_anno_file1 = f'{root}/test_label.txt'
test = dict(
type=dataset_type,
img_prefix=f'{root}/imgs/test',
ann_file=test_anno_file1,
loader=dict(
type='AnnFileLoader',
repeat=1,
file_format='txt',
file_storage_backend='disk',
parser=dict(type='LineJsonParser', keys=['filename', 'text'])),
pipeline=None,
test_mode=True)

train_list = [train1]

test_list = [test]

train_label:
word_820.png LONG

test_label:{"filename": "word_9.png","text":"SONY"}

I didn't find the reason why the model wasn't trained properly, hoping to get some find the reason for the error