SHAPE : Self-Improved Visual Preference Alignment by Iteratively Generating Holistic Winner
This addresses the challenge of reliable alignment for LVLMs in vision-language tasks, offering a cheaper and more effective alternative to human annotations, though it is incremental as it builds on existing preference fine-tuning methods.
The paper tackles the problem of limited diversity and high costs in human-annotated preference data for aligning Large Visual Language Models (LVLMs) by introducing SHAPE, a self-supervised framework that transforms supervised text-image pairs into holistic preference triplets, achieving gains such as +11.3% on MMVet and +8.0% on POPE over baselines in 7B models.
Large Visual Language Models (LVLMs) increasingly rely on preference alignment to ensure reliability, which steers the model behavior via preference fine-tuning on preference data structured as ``image - winner text - loser text'' triplets. However, existing approaches often suffer from limited diversity and high costs associated with human-annotated preference data, hindering LVLMs from fully achieving their intended alignment capabilities. We present \projectname, a self-supervised framework capable of transforming the already abundant supervised text-image pairs into holistic preference triplets for more effective and cheaper LVLM alignment, eliminating the need for human preference annotations. Our approach facilitates LVLMs in progressively enhancing alignment capabilities through iterative self-improvement. The key design rationale is to devise preference triplets where the winner text consistently improves in holisticness and outperforms the loser response in quality, thereby pushing the model to ``strive to the utmost'' of alignment performance through preference fine-tuning. For each given text-image pair, SHAPE introduces multiple visual augmentations and pairs them with a summarized text to serve as the winner response, while designating the original text as the loser response. Experiments across \textbf{12} benchmarks on various model architectures and sizes, including LLaVA and DeepSeek-VL, show that SHAPE achieves significant gains, for example, achieving +11.3\% on MMVet (comprehensive evaluation), +1.4\% on MMBench (general VQA), and +8.0\% on POPE (hallucination robustness) over baselines in 7B models. Notably, qualitative analyses confirm enhanced attention to visual details and better alignment with human preferences for holistic descriptions.