RICO: Improving Accuracy and Completeness in Image Recaptioning via Visual Reconstruction

TL;DR

RICO introduces a visual reconstruction-based framework that iteratively refines image captions to improve their accuracy and completeness, addressing hallucination and missing detail issues in existing methods.

Contribution

The paper presents RICO, a novel visual reconstruction approach that enhances image captioning by iteratively identifying discrepancies and refining descriptions, with a lightweight variant RICO-Flash for efficiency.

Findings

Significantly improves caption accuracy and completeness.

Outperforms most baselines by approximately 10% on CapsBench and CompreCap.

Demonstrates effectiveness of visual reconstruction in caption refinement.

Abstract

Image recaptioning is widely used to generate training datasets with enhanced quality for various multimodal tasks. Existing recaptioning methods typically rely on powerful multimodal large language models (MLLMs) to enhance textual descriptions, but often suffer from inaccuracies due to hallucinations and incompleteness caused by missing fine-grained details. To address these limitations, we propose RICO, a novel framework that refines captions through visual reconstruction. Specifically, we leverage a text-to-image model to reconstruct a caption into a reference image, and prompt an MLLM to identify discrepancies between the original and reconstructed images to refine the caption. This process is performed iteratively, further progressively promoting the generation of more faithful and comprehensive descriptions. To mitigate the additional computational cost induced by the iterative process, we introduce RICO-Flash, which learns to generate captions like RICO using DPO. Extensive experiments demonstrate that our approach significantly improves caption accuracy and completeness, outperforms most baselines by approximately 10% on both CapsBench and CompreCap. Code released at https://github.com/wangyuchi369/RICO.