01-convex_vs_concave

Bongard Problem: Convex VS Concave

Introduction

We have manually designed more than 600 shapes with different attributes for Bongard problem synthesis. This demon shows how to take advantage of the human designed shape information to generate Bongard problems. In particular, we created a "Convex VS Concave" Bongard problem as an example.

Note that all the negative images does not necessarily have to form a concept. Instead, each of them violating the concept in positive images would be sufficient for a Bongard problem.

Usages

To generate the Bongard demo images, please run the following command in the terminal.

python demo.py

This Python script will read the human_designed_shapes.tsv and human_designed_shapes_attributes.tsv provided by the Bongard library for sampling shapes and generate OneStrokeShape. The users could also design their own shapes and attributes and put them into the two tsv files.

The generated vector images would be saved in demo/ps, whereas the corresponding PNG images would be saved in demo/png. There is also a merged image from PNG images for human visualization in demo.