Self-Improving VLM Judges Without Human Annotations

TL;DR

This paper introduces a novel self-training framework for VLM judges that eliminates the need for costly human annotations by using self-synthesized data, leading to significant performance improvements across multiple benchmarks.

Contribution

The work presents an iterative self-training method for VLM judges that leverages self-synthesized data, outperforming larger models without human preference annotations.

Findings

Improved judge accuracy from 0.38 to 0.51 on VL-RewardBench.

Outperforms larger models like Llama-3.2-90B, GPT-4o, and Claude 3.5 Sonnet.

Strong gains in general, hallucination, and reasoning dimensions.

Abstract

Effective judges of Vision-Language Models (VLMs) are crucial for model development. Current methods for training VLM judges mainly rely on large-scale human preference annotations. However, such an approach is costly, and the annotations easily become obsolete as models rapidly improve. In this work, we present a framework to self-train a VLM judge model without any human preference annotations, using only self-synthesized data. Our method is iterative and has three stages: (1) generate diverse multimodal instruction-response pairs at varying quality levels, (2) generate reasoning traces and judgments for each pair, removing the ones that do not match our expected quality levels, and (3) training on correct judge answers and their reasoning traces. We evaluate the resulting judge on Multimodal RewardBench and VL-RewardBench across domains: correctness, preference, reasoning, safety, and visual question-answering. Our method improves a Llama-3.2-11B multimodal judge from 0.38 to 0.51 in overall accuracy on VL-RewardBench, often outperforming much larger models including Llama-3.2-90B, GPT-4o, and Claude 3.5 Sonnet, with particularly strong gains in general, hallucination, and reasoning dimensions. The overall strength of these human-annotation-free results suggest the potential for a future self-judge that evolves alongside rapidly improving VLM capabilities.