This implementation requires the input data in the following format:
txt_file
txt file: multi-line with format:ImageFileName,TextLabel
. e.g.data/example_gt.txt
img_root
folder: image files corresponding toImageFileName
. e.g.data/example_imgs
img_folder
folder: image files.
Or use lmdb format dataset and modified the corresponding config file.
Base on lmdb dataset in deep text recognition benchmark
: Source repository
Download lmdb dataset: Here
Unzip and modify folder below and use config file config_lmdb.json
to train:
(Can custom sub folder in folder training
and val
by fix select_data
in config file)
dataset/data_lmdb_release/
training/
MJ_train
ST
val/
MJ_test
MJ_valid