Simulated data for bangali text localisation in natural image

(1) The dataset comprises paired images, consisting of 2D simulated text on the styled image (i_s), text image (i_t), masking of text (mask_t), real background image (t_b), real sample image (t_f), text skeleton (t_sk), and text styled image (t_t).
(2) While generating synthetic data, we scale the text image in our experiment to a height of 64 pixels while maintaining the original aspect ratio.
(3) The source dataset has background images, color, fonts, and text, whereas the synthetic dataset contains 49,000 random images.
(4) The batch size is set to 8, and the input images are resized in size to approximately W×64 while maintaining the aspect ratio.