peal.data.dataset_generators.SparseNumbersDatasetGenerator

class peal.data.dataset_generators.SparseNumbersDatasetGenerator

Bases: object

Generates a dataset of images with a 2x4 grid of 4-digit numbers. Resolution: 128x128. Grid: 2 columns, 4 rows. Numbers: 0000-9999, each appears 50 times in the dataset. Occupancy: Each slot has a 50% chance of being present (500,000 total numbers in 125,000 samples). Background: Variable red intensity (white to red). Foreground: Black numbers.

Methods

generate_dataset

Generates the sparse_numbers dataset.

__init__(data_config, **kwargs)

Parameters:

• data_config (DataConfig) – Supplies dataset_path, num_samples (default 125000) and output_split (number of distinct numbers, default 10000).

• **kwargs – font_path overrides the DejaVu Sans TrueType font.

generate_dataset()

Generates the sparse_numbers dataset.

Writes imgs/<i:06d>.png and data.csv with columns imgs, Red1..Red8 (background red intensity per slot) and Num1..Num8 (number per slot, -1 if empty) to data_config.dataset_path; an existing folder is moved aside with a timestamp suffix. Uses the global random module (unseeded).