LogicRank: Logic Induced Reranking for Generative Text-to-Image Systems

TL;DR

LogicRank is a neuro-symbolic framework designed to improve the accuracy of reranking generated images from text-to-image models by enhancing logical precision and consistency.

Contribution

It introduces LogicRank, a novel neuro-symbolic reasoning method that enhances reranking accuracy and can be integrated into existing text-to-image generation workflows.

Findings

LogicRank outperforms CLIP in reranking accuracy.

State-of-the-art models struggle with precise statement generation.

LogicRank can be used to fine-tune models for better logical consistency.

Abstract

Text-to-image models have recently achieved remarkable success with seemingly accurate samples in photo-realistic quality. However as state-of-the-art language models still struggle evaluating precise statements consistently, so do language model based image generation processes. In this work we showcase problems of state-of-the-art text-to-image models like DALL-E with generating accurate samples from statements related to the draw bench benchmark. Furthermore we show that CLIP is not able to rerank those generated samples consistently. To this end we propose LogicRank, a neuro-symbolic reasoning framework that can result in a more accurate ranking-system for such precision-demanding settings. LogicRank integrates smoothly into the generation process of text-to-image models and moreover can be used to further fine-tune towards a more logical precise model.