Quizzard@INOVA Challenge 2025 -- Track A: Plug-and-Play Technique in Interleaved Multi-Image Model

TL;DR

This paper evaluates a multi-modal interleaved reasoning model across diverse datasets, demonstrating the effectiveness of plug-and-play enhancements like DCI in improving semantic and structured understanding.

Contribution

It introduces the LLaVA-NeXT-interleave model and the DCI connector, showing their performance across multiple multi-image reasoning tasks and datasets.

Findings

Standard model excels in vision-heavy tasks.

DCI-enhanced model performs better on semantic coherence datasets.

Plug-and-play techniques improve multi-modal reasoning capabilities.

Abstract

This paper addresses two main objectives. Firstly, we demonstrate the impressive performance of the LLaVA-NeXT-interleave on 22 datasets across three different tasks: Multi-Image Reasoning, Documents and Knowledge-Based Understanding and Interactive Multi-Modal Communication. Secondly, we add the Dense Channel Integration (DCI) connector to the LLaVA-NeXT-Interleave and compare its performance against the standard model. We find that the standard model achieves the highest overall accuracy, excelling in vision-heavy tasks like VISION, NLVR2, and Fashion200K. Meanwhile, the DCI-enhanced version shows particular strength on datasets requiring deeper semantic coherence or structured change understanding such as MIT-States_PropertyCoherence and SlideVQA. Our results highlight the potential of combining powerful foundation models with plug-and-play techniques for Interleave tasks. The code is available at https://github.com/dinhvietcuong1996/icme25-inova.