This code is supposed to use SROIE dataset for training the model to detect text from weld-map images correctly.
the dataset can be downloaded from this link, place the train and test folders in raw/sroie directory. and run
python dataset.py
to convert the dataset into required format, it will produce two folders data/processed/img
and data/processed/gt
move these folders to
src/htr/data
and then move to training process.
For Model Training run
python main.py --mode train --data_dir htr/data
Remeber to place the model inside the src/htr/model directory.
There are two directories currently.
src/htr/model
-> Contains model trained by us.src/htr/model_temp
-> Contains model provided by HTR repository
To generate a report of text recognition from the available weldmap images run, If want to generate report through the model
provided by author place content of model_temp
inside model
python main.py --mode gen_report
The dataset link has been emailed to you from id [email protected]
Place the Monsonato.pdf inside the raw
directory
NOTE: Data is to be labeled from hand for report generation.
Place the generated images inside the data/weldmaps
directory to generate report.