Token-level Correlation-guided Compression for Efficient Multimodal Document Understanding

Cropping high-resolution document images into multiple sub-images is the most widely used approach for current Multimodal Large Language Models (MLLMs) to do document understanding. Most of current document understanding methods preserve all tokens within sub-images and treat them equally. This neglects their different informativeness and leads to a significant increase in the number of image tokens. To perform a more adaptive and efficient document understanding, we propose Token-level Correlation-guided Compression, a parameter-free and plug-and-play methodology to optimize token processing. Firstly, we propose an innovative approach for assessing the pattern repetitiveness based on the correlation between each patch tokens. This method identifies redundant tokens, allowing for the determination of the sub-image's information density. Secondly, we present a token-level sampling method that efficiently captures the most informative tokens by delving into the correlation between the [CLS] token and patch tokens. By integrating these strategies, we develop a plug-and-play adaptive compressor module that can be seamlessly incorporated into MLLMs utilizing cropping techniques. This module not only enhances the processing speed during training and inference but also maintains comparable performance. We conduct experiments with the SOTA document understanding model mPLUG-DocOwl1.5 and the effectiveness is demonstrated through extensive comparisons with other compression methods.

A Novel Dual Quaternion Based Dynamic Motion Primitives for Acrobatic Flight

The realization of motion description is a challenging work for fixed-wing Unmanned Aerial Vehicle (UAV) acrobatic flight, due to the inherent coupling problem in ranslational-rotational motion. This paper aims to develop a novel maneuver description method through the idea of imitation learning, and there are two main contributions of our work: 1) A dual quaternion based dynamic motion primitives (DQ-DMP) is proposed and the state equations of the position and attitude can be combined without loss of accuracy. 2) An online hardware-inthe-loop (HITL) training system is established. Based on the DQDMP method, the geometric features of the demonstrated maneuver can be obtained in real-time, and the stability of the DQ-DMP is theoretically proved. The simulation results illustrate the superiority of the proposed method compared to the traditional position/attitude decoupling method.