Doc-CoB: Enhancing Multi-Modal Document Understanding with Visual Chain-of-Boxes Reasoning

Multimodal large language models (MLLMs) have made significant progress in document understanding. However, the information-dense nature of document images still poses challenges, as most queries depend on only a few relevant regions, with the rest being redundant. Existing one-pass MLLMs process entire document images without considering query relevance, often failing to focus on critical regions and producing unfaithful responses. Inspired by the human coarse-to-fine reading pattern, we introduce Doc-CoB (Chain-of-Box), a simple-yet-effective mechanism that integrates human-style visual reasoning into MLLM without modifying its architecture. Our method allows the model to autonomously select the set of regions (boxes) most relevant to the query, and then focus attention on them for further understanding. We first design a fully automatic pipeline, integrating a commercial MLLM with a layout analyzer, to generate 249k training samples with intermediate visual reasoning supervision. Then we incorporate two enabling tasks that improve box identification and box-query reasoning, which together enhance document understanding. Extensive experiments on seven benchmarks with four popular models show that Doc-CoB significantly improves performance, demonstrating its effectiveness and wide applicability. All code, data, and models will be released publicly.

Local track irregularity identification based on multi-sensor time-frequency features of high-speed railway bridge accelerations

Shortwave track diseases are generally reflected in the form of local track irregularity. Such diseases will greatly impact the train-track-bridge interaction (TTBI) dynamic system, seriously affecting train safety. Therefore, a method is proposed to detect and localize local track irregularities based on multis-sensor time-frequency features of high-speed railway bridge accelerations. Continuous wavelet transform (CWT) is used to analyze the multi-sensor accelerations of railway bridges. Moreover, time-frequency features based on the sum of wavelet coefficients are proposed, considering the influence of the distance from the measurement points to the local irregularity on the recognition accuracy. Then, the multi-domain features are utilized to recognize deteriorated railway locations. A simply-supported high-speed railway bridge traversed by a railway train is adopted as a numerical simulation. Comparative studies are conducted to investigate the influence of vehicle speeds and the location of local track irregularity on the algorithm. Numerical simulation results show that the proposed algorithm can detect and locate local track irregularity accurately and is robust to vehicle speeds.