These are the Python scripts used for generating the Bongard-LOGO dataset. The Bongard-Logo dataset consists of three types of Bongard problems, freeform, basic, and abstract. For more details about the design and instructions on these three types of Bongard problems, please read the Bongard-Logo dataset documentation.
To generate the dataset, please run the following command in the terminal.
python run.pyOnce the generation is complete, there should be 3600 freeform problems, 4000 basic problems, and 4400 abstract problems.
$ ls images/ff/png/ | wc -l
3600
$ ls images/bd/png/ | wc -l
4000
$ ls images/hd/png/ | wc -l
4400
The action programs for each type of the problems are saved as ff_action_program, bd_action_program.json, hd_action_program.json, respectively, in each directory.