VERSE: Visual Embedding Reduction and Space Exploration. Clustering-Guided Insights for Training Data Enhancement in Visually-Rich Document Understanding
This work addresses performance enhancement for Visually-Rich Document Understanding models, offering a method to optimize on-premise solutions, though it is incremental as it builds on existing embedding analysis techniques.
The paper tackles the problem of improving Vision-Language Models for Visually-Rich Document Understanding by analyzing visual embedding spaces to identify error-prone clusters and generate synthetic data for retraining. Results show that this approach boosts F1 performance without degrading generalization and enables on-premise models to match or surpass SaaS solutions like GPT-4 and Pixtral.
This work introduces VERSE, a methodology for analyzing and improving Vision-Language Models applied to Visually-rich Document Understanding by exploring their visual embedding space. VERSE enables the visualization of latent representations, supporting the assessment of model feasibility. It also facilitates the identification of problematic regions and guides the generation of synthetic data to enhance performance in those clusters. We validate the methodology by training on the synthetic MERIT Dataset and evaluating on its real-world counterpart, MERIT Secret. Results show that VERSE helps uncover the visual features associated with error-prone clusters, and that retraining with samples containing these features substantially boosts F1 performance without degrading generalization. Furthermore, we demonstrate that on-premise models such as Donut and Idefics2, when optimized with VERSE, match or even surpass the performance of SaaS solutions like GPT-4 and Pixtral.