Abstract Visual Reasoning with Tangram Shapes

TL;DR

KiloGram is a large, richly annotated dataset designed to evaluate and improve abstract visual reasoning in humans and machines, highlighting the benefits of part-based descriptions and fine-tuning for model performance.

Contribution

The paper introduces KiloGram, a novel, extensive dataset with detailed annotations for studying abstract visual reasoning, and demonstrates how part-based descriptions enhance reasoning capabilities.

Findings

Pre-trained models show limited reasoning, improved with fine-tuning.

Explicit part descriptions aid reasoning for humans and models.

Joint encoding of visual and linguistic inputs enhances performance.

Abstract

We introduce KiloGram, a resource for studying abstract visual reasoning in humans and machines. Drawing on the history of tangram puzzles as stimuli in cognitive science, we build a richly annotated dataset that, with >1k distinct stimuli, is orders of magnitude larger and more diverse than prior resources. It is both visually and linguistically richer, moving beyond whole shape descriptions to include segmentation maps and part labels. We use this resource to evaluate the abstract visual reasoning capacities of recent multi-modal models. We observe that pre-trained weights demonstrate limited abstract reasoning, which dramatically improves with fine-tuning. We also observe that explicitly describing parts aids abstract reasoning for both humans and models, especially when jointly encoding the linguistic and visual inputs. KiloGram is available at https://lil.nlp.cornell.edu/kilogram .