Search papers, labs, and topics across Lattice.
The paper introduces Image-Only Training for UMMs (IOMM), a two-stage framework to improve the efficiency of pre-training visual generation components in Unified Multimodal Models (UMMs). IOMM decouples the visual generation pre-training from paired text-image data by pre-training exclusively on unlabeled images, followed by fine-tuning on a mixture of images and a small set of paired data. Experiments show that IOMM achieves state-of-the-art performance on GenEval and WISE benchmarks with significantly reduced training costs (e.g., IOMM-B (3.6B) trained in ~1050 H800 GPU hours).
Train your UMM visual generation component on image-only data first and you'll get SOTA performance with a fraction of the compute.
Unified Multimodal Models (UMMs) are often constrained by the pre-training of their $\textbf{visual generation components}$, which typically relies on inefficient paradigms and scarce, high-quality text-image paired data. In this paper, we systematically analyze pre-training recipes for $\textbf{UMM visual generation}$ and identify these two issues as the major bottlenecks. To address them, we propose $\textbf{Image-Only Training for UMMs (IOMM)}$, a data-efficient two-stage training framework. The first stage pre-trains the visual generative component $\textbf{exclusively}$ using abundant unlabeled image-only data, thereby removing the dependency on paired data $\textbf{for this costly phase}$. The second stage fine-tunes the model using a mixture of unlabeled images and a small curated set of text-image pairs, leading to improved instruction alignment and generative quality. Extensive experiments show that IOMM not only improves training efficiency but also achieves state-of-the-art (SOTA) performance. For example, our IOMM-B (3.6B) model was trained from scratch using only $\sim \textbf{1050}$ H800 GPU hours (with the vast majority, $\textbf{1000}$ hours, dedicated to the efficient $\textbf{image-only pre-training stage}$). It achieves $\textbf{0.89}$ on GenEval and $\textbf{0.55}$ on WISE--surpassing strong baselines such as BAGEL-7B (0.82&0.55) and BLIP3-o-4B (0.84&0.50). Code is available $\href{https://github.com/LINs-lab/IOMM}{https://github.com/LINs-lab/IOMM}$.
