Yuchuan Deng

Jiazhen Liu, Yuchuan Deng, Long Chen

Empowering Small-scale Vision-Language Models (SVLMs) with reliable thinking capabilities remains fundamentally challenging due to their limited parameter capacity and weak instruction-following abilities. Existing training paradigms, including Supervised Fine-Tuning (SFT) and Reinforcement Learning with Verifiable Reward (RLVR), impose substantial demands on the base VLM, exceeding the capabilities of SVLMs. Consequently, directly applying these paradigms to SVLMs often suffers from severe pseudo thinking traces and advantage collapse, ultimately undermining both thinking reliability and task performance. A natural solution is to combine SFT and RLVR, leveraging their complementarity to reduce the dependence on model capacity. However, the widely adopted two-stage training paradigm still performs poorly on SVLMs, as their tendency toward sub-optimal convergence hinders the trade-off and limits the benefits of the combination. To address this, we propose DyME, a novel training paradigm that Dynamically selects between Memorization (via SFT) and Exploration (via RLVR) modes at each optimization step, ensuring that every update contributes to the trade-off. Extensive experiments across diverse domains demonstrate that DyME consistently achieves this balance, and thus delivers substantial performance improvements. These results establish DyME as a practical and effective solution for empowering SVLMs with reliable thinking capabilities. GitHub: https://github.com/HKUST-LongGroup/DyME

Yuchuan Deng, Qijie Wei, Kaiheng Qian et al.

Fundus imaging such as CFP, OCT and UWF is crucial for the early detection of retinal anomalies and diseases. Fundus image understanding, due to its knowledge-intensive nature, poses a challenging vision-language task. An emerging approach to addressing the task is to post-train a generic multimodal large language model (MLLM), either by supervised finetuning (SFT) or by reinforcement learning with verifiable rewards (RLVR), on a considerable amount of in-house samples paired with high-quality clinical reports. However, these valuable samples are not publicly accessible, which not only hinders reproducibility but also practically limits research to few players. To overcome the barrier, we make a novel attempt to train a reasoning-enhanced fundus-reading MLLM, which we term Fundus-R1, using exclusively public datasets, wherein over 94\% of the data are annotated with only image-level labels. Our technical contributions are two-fold. First, we propose a RAG-based method for composing image-specific, knowledge-aware reasoning traces. Such auto-generated traces link visual findings identified by a generic MLLM to the image labels in terms of ophthalmic knowledge. Second, we enhance RLVR with a process reward that encourages self-consistency of the generated reasoning trace in each rollout. Extensive experiments on three fundus-reading benchmarks, i.e., FunBench, Omni-Fundus and GMAI-Fundus, show that Fundus-R1 clearly outperforms multiple baselines, including its generic counterpart (Qwen2.5-VL) and a stronger edition post-trained without using the generated traces. This work paves the way for training powerful fundus-reading MLLMs with publicly available data.

Yuchuan Deng, Zhanpeng Hu, Jiakun Han et al.

Text-based Person Re-identification (TPR) aims to retrieve specific individual images from datasets based on textual descriptions. Existing TPR methods primarily focus on recognizing explicit and positive characteristics, often overlooking the role of negative descriptions. This oversight can lead to false positives-images that meet positive criteria but should be excluded based on negative descriptions. To address these limitations, we introduce DualFocus, a unified framework that integrates plausible descriptions to enhance the interpretative accuracy of vision-language models in TPR tasks. DualFocus leverages Dual (Positive/Negative) Attribute Prompt Learning (DAPL), which incorporates Dual Image-Attribute Contrastive (DIAC) Learning and Sensitive Image-Attributes Matching (SIAM) Learning, enabling the detection of non-existent attributes and reducing false positives. To achieve a balance between coarse and fine-grained alignment of visual and textual embeddings, we propose the Dynamic Tokenwise Similarity (DTS) loss, which refines the representation of both matching and non-matching descriptions, thereby improving the matching process through detailed and adaptable similarity assessments. The comprehensive experiments on CUHK-PEDES, ICFG-PEDES, and RSTPReid, DualFocus demonstrates superior performance over state-of-the-art methods, significantly enhancing both precision and robustness in TPR.