Search papers, labs, and topics across Lattice.
This paper introduces HIVE, a hierarchical pre-training framework that enhances vision-language alignment by incorporating hierarchical cross-attention between vision encoders and LLMs. HIVE enables structured feature fusion across multiple layers of the vision encoder, improving gradient flow and representation learning through a novel three-stage training strategy. Experiments show HIVE outperforms existing methods on image classification and vision-language tasks like MME, GQA, OK-VQA, and ScienceQA, demonstrating the benefits of hierarchical feature integration.
Hierarchical cross-attention between vision encoders and LLMs unlocks superior vision-language alignment, outperforming flattened feature fusion methods on a range of benchmarks.
The field of computer vision has experienced significant advancements through scalable vision encoders and multimodal pre-training frameworks. However, existing approaches often treat vision encoders and large language models (LLMs) as independent modules, limiting the integration of hierarchical visual features. In this work, we propose HIVE (Hierarchical Pre-Training of Vision Encoders), a novel framework that enhances vision-language alignment by introducing hierarchical cross-attention between the vision encoder and LLM. Unlike conventional methods that flatten image embeddings, HIVE enables structured feature fusion across multiple layers, improving gradient flow and representation learning. To optimize this interaction, we introduce a three-stage training strategy that progressively aligns the vision encoder with the LLM, ensuring stable optimization and effective multimodal fusion. Empirical evaluations demonstrate that HIVE achieves superior performance not only in image classification but also on various vision-language tasks, outperforming self-attention-based methods in benchmarks such as MME, GQA, OK-VQA, and ScienceQA. Our results highlight the benefits of hierarchical feature integration, paving the way for more efficient and expressive vision-language models.
