Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2512.05145 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914180350607360 |
|---|---|
| author | Lin, Inna Wanyin Hu, Yushi Li, Shuyue Stella Geng, Scott Koh, Pang Wei Zettlemoyer, Luke Althoff, Tim Ghazvininejad, Marjan |
| author_facet | Lin, Inna Wanyin Hu, Yushi Li, Shuyue Stella Geng, Scott Koh, Pang Wei Zettlemoyer, Luke Althoff, Tim Ghazvininejad, Marjan |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2512_05145 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Self-Improving VLM Judges Without Human Annotations Lin, Inna Wanyin Hu, Yushi Li, Shuyue Stella Geng, Scott Koh, Pang Wei Zettlemoyer, Luke Althoff, Tim Ghazvininejad, Marjan Computer Vision and Pattern Recognition 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. |
| title | Self-Improving VLM Judges Without Human Annotations |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2512.05145 |