To start from scratch we need to generate training and testing samples for the model.
If we wish to generate more samples the code is provided on datagen folder and we should run python ./datagen/draw_plot.py. This file takes input arguments to change the number of instances to generate listed below.
--cnt_bezier: The number of bezier curve samples can be changed using this argument.--cnt_scatter: The number of scatter plot samples can be changed using this argument.--cnt_polygon: The number of polygon samples can be changed using this argument.--p_figsize: This argument controls the percentage of squared, vertical and horizontal plots. For instance with [0.5, 0.3, 0.2] results in 50% squared, 30% vertical and 20% horizontal plots.--p_1D: This argument controls the percentage of bezier curves that strictly have no intersections.--p_grid: This argument controls the percentage of samples with gridlines.
As an example of running the code using all of these arguments we can write python ./datagen/draw_plot.py --cnt_bezier 10 --cnt_scatter 5 --cnt_polygon 6 --p_figsize 0.33 0.33 0.33 --p_1D 0.9 --p_grid 0.1. After generating samples, we should split them as train and test sets each having source and tactile folders.
NOTE: By default a dataset of 5000 samples (2000 bezier curves, 1500 polygons and, 1500 scatter plots) has been provided. 90% of the data was assigned to training process and the rest of them belongs to test process. Both of the subsets are balanced and we can find them on data directory.