Pix2Code: Learning to Compose Neural Visual Concepts as Programs

TL;DR

Pix2Code introduces a hybrid framework combining symbolic and neural methods to learn and interpret visual relational concepts as programs, enabling better generalization and human interpretability.

Contribution

It extends program synthesis to visual reasoning, integrating symbolic and neural representations for interpretable and generalizable visual concept learning.

Findings

Pix2Code effectively identifies compositional visual concepts.

The framework generalizes well to novel data and configurations.

Representations are human interpretable and easily revisable.

Abstract

The challenge in learning abstract concepts from images in an unsupervised fashion lies in the required integration of visual perception and generalizable relational reasoning. Moreover, the unsupervised nature of this task makes it necessary for human users to be able to understand a model's learnt concepts and potentially revise false behaviours. To tackle both the generalizability and interpretability constraints of visual concept learning, we propose Pix2Code, a framework that extends program synthesis to visual relational reasoning by utilizing the abilities of both explicit, compositional symbolic and implicit neural representations. This is achieved by retrieving object representations from images and synthesizing relational concepts as lambda-calculus programs. We evaluate the diverse properties of Pix2Code on the challenging reasoning domains, Kandinsky Patterns and CURI, thereby testing its ability to identify compositional visual concepts that generalize to novel data and concept configurations. Particularly, in stark contrast to neural approaches, we show that Pix2Code's representations remain human interpretable and can be easily revised for improved performance.