Search papers, labs, and topics across Lattice.
The paper addresses the challenge of scaling visual preference optimization due to noisy and limited preference datasets. They introduce Poly-DPO, a modified DPO objective with a polynomial term to dynamically adjust model confidence, enhancing robustness against noise in preference data. They also construct ViPO, a large-scale, high-resolution visual preference dataset with 1M image pairs and 300K video pairs, demonstrating that high-quality data allows standard DPO to achieve state-of-the-art results, while Poly-DPO is beneficial for noisy datasets.
Scaling visual preference optimization hinges on data quality, as demonstrated by the finding that standard DPO suffices for a sufficiently large and clean dataset, while a novel Poly-DPO objective is crucial for noisy data.
While preference optimization is crucial for improving visual generative models, how to effectively scale this paradigm remains largely unexplored. Current open-source preference datasets contain conflicting preference patterns, where winners excel in some dimensions but underperform in others. Naively optimizing on such noisy datasets fails to learn preferences, hindering effective scaling. To enhance robustness against noise, we propose Poly-DPO, which extends the DPO objective with an additional polynomial term that dynamically adjusts model confidence based on dataset characteristics, enabling effective learning across diverse data distributions. Beyond biased patterns, existing datasets suffer from low resolution, limited prompt diversity, and imbalanced distributions. To facilitate large-scale visual preference optimization by tackling data bottlenecks, we construct ViPO, a massive-scale preference dataset with 1M image pairs at 1024px across five categories and 300K video pairs at 720p+ across three categories. State-of-the-art generative models and diverse prompts ensure reliable preference signals with balanced distributions. Remarkably, when applying Poly-DPO to our high-quality dataset, the optimal configuration converges to standard DPO. This convergence validates dataset quality and Poly-DPO's adaptive nature: sophisticated optimization becomes unnecessary with sufficient data quality, yet remains valuable for imperfect datasets. We validate our approach across visual generation models. On noisy datasets like Pick-a-Pic V2, Poly-DPO achieves 6.87 and 2.32 gains over Diffusion-DPO on GenEval for SD1.5 and SDXL, respectively. For ViPO, models achieve performance far exceeding those trained on existing open-source preference datasets. These results confirm that addressing both algorithmic adaptability and data quality is essential for scaling visual preference optimization.
