ExpertVerse: A General-Purpose Benchmark for Expert-Level Reasoning in Knowledge-Intensive Visual Synthesis

2026-07-21

AI Summary

This work addresses the limited expert-level reasoning capabilities of existing vision-generation models in knowledge-intensive tasks. To this end, we introduce ExpertVerse, a novel benchmark that establishes an orthogonal evaluation framework spanning nine cognitive abilities and eight expert domains, accompanied by ExpertVerse-100K—a large-scale dataset featuring reasoning trajectories and knowledge-anchored explanations. Building upon this foundation, we develop KnowThinker, a new model that integrates reinforcement learning fine-tuning with Bootstrapped Reward Rectification (BRR) and Conflict-aware Pareto Advantage Fusion (CPAF), yielding a multi-objective optimization algorithm termed BPPO. Extensive experiments uncover critical shortcomings in prevailing models and demonstrate KnowThinker’s significant superiority in complex knowledge-driven reasoning tasks.

Abstract

Recent advances in multimodal generative models have enabled instruction-based image generation to move beyond semantic manipulation to knowledge-driven visual reasoning. However, these methods focus on explicit commonsense reasoning, shallow causal understanding, and direct knowledge recall, failing at knowledge-intensive generation. We develop \textbf{ExpertVerse}, a capability-centric benchmark to evaluate generative models via knowledge-intensive lens. ExpertVerse stratifies reasoning generation across an orthogonal taxonomy of \textit{9 cognitive capabilities} and \textit{8 expert disciplines}, yielding \textit{58 sub-disciplines}. We curate 1,611 expert-annotated instances covering single-image editing, multi-image composition, and text-to-image generation. We further develop an automated workflow to produce \textbf{ExpertVerse-100K}, a large-scale dataset with reasoning traces and knowledge-anchored rationale annotations. Based on this, we train \textbf{KnowThinker} with RL fine-tuning, a VLM reasoning engine with world knowledge that jointly generates thinking processes and refined instructions. Towards the cross-modal credit misalignment and multi-objective gradient conflicts in multi-reward optimization, we propose a tailored Bootstrapped Pareto Policy Optimization (BPPO), which synergizes Bootstrapping Reward Rectification (BRR) and Conflict-Aware Pareto Advantage Fusion (CPAF). Extensive results of both open-source and proprietary models exposes critical reasoning deficits, highlighting imperative for knowledge-intensive benchmarks towards next-generation visual generation.

Problem

• knowledge-intensive reasoning
• visual synthesis
• expert-level reasoning
• multimodal generation
• reasoning benchmark

Innovation

• knowledge-intensive reasoning
• multimodal generative benchmark
• reasoning trace annotation
• reinforcement learning fine-tuning
• multi-objective optimization