Weld-Maps Table Recognition

1. Dataset Preparation

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.

2. Model Training

For Model Training run

python main.py --mode train --data_dir htr/data

3. Model Evaluation

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

4. Extracting Images from Weld-Maps dataset

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.