Wang Song

Yifeng Wu, Huimin Huang, Ruiluo Wu et al.

Multimodal Large Language Models (MLLMs) often lose track of the right image regions during fine-grained spatial reasoning, because a textual query rarely carries any explicit geometric anchor into the pixel domain. Prevailing remedies either rewire the model's weights or pad the prompt with verbose instructions, yet neither reliably pins the language to the correct visual coordinates without eroding the backbone's general competence. We introduce Timage, a paradigm that recasts multimodal understanding as an alignment problem solved at the input: the query is drawn, as a typeset overlay, onto the image itself. The placement and appearance of this overlay are produced by a Constrained Schrödinger Bridge (cSB), an entropic optimal-transport sampler that factorizes layout synthesis into two coupled stochastic stages. The first stage, Region Search, transports noise toward query-aligned image zones while obeying a hard occlusion barrier that protects salient foreground content; the second stage, Appearance Shaping, sizes the glyphs through an ``ink-budget'' regularizer so that the rendered text stays legible and visually balanced. The resulting overlay behaves as an explicit attention beacon that channels the model's focus along spatial semantics. On the VMCBench suite, Timage paired with a modest 7B backbone clearly overtakes far larger proprietary systems as well as parameter-tuned baselines. The study positions deliberate input reconstruction as a powerful, architecture-neutral lever for strengthening multimodal reasoning.

Huang Yuchao, Wei Moshi, Wang Song et al.

Code comment generation techniques aim to generate natural language descriptions for source code. There are two orthogonal approaches for this task, i.e., information retrieval (IR) based and neural-based methods. Recent studies have focused on combining their strengths by feeding the input code and its similar code snippets retrieved by the IR-based approach to the neural-based approach, which can enhance the neural-based approach's ability to output low-frequency words and further improve the performance. However, despite the tremendous progress, our pilot study reveals that the current combination is not generalizable and can lead to performance degradation. In this paper, we propose a straightforward but effective approach to tackle the issue of existing combinations of these two comment generation approaches. Instead of binding IR- and neural-based approaches statically, we combine them in a dynamic manner. Specifically, given an input code snippet, we first use an IR-based technique to retrieve a similar code snippet from the corpus. Then we use a Cross-Encoder based classifier to decide the comment generation method to be used dynamically, i.e., if the retrieved similar code snippet is a true positive (i.e., is semantically similar to the input), we directly use the IR-based technique. Otherwise, we pass the input to the neural-based model to generate the comment. We evaluate our approach on a large-scale dataset of Java projects. Experiment results show that our approach can achieve 25.45 BLEU score, which improves the state-of-the-art IR-based approach, neural-based approach, and their combination by 41%, 26%, and 7%, respectively. We propose a straightforward but effective dynamic combination of IR-based and neural-based comment generation, which outperforms state-of-the-art approaches by a substantial margin.