Jintao Tong

Jinghan Zhao, Wenwei Jin, Anqi Li et al.

Item-to-Item (I2I) retrieval is a fundamental part of modern content platforms, supporting critical industrial workflows from recommendation engines to content auditing. While multimodal embedding methods have advanced general retrieval, they often falter in I2I scenarios due to the challenges of balancing global content representation with fine-grained local retrieval, the systemic inefficiency of decoupled embedding-and-ranking pipelines, and the inherent trade-offs between model precision and serving latency. To solve these issues, we propose \textbf{UniNote}, a unified embedding model designed for industrial I2I retrieval. Tailored retrieval strategies are introduced to support representation learning over complex, multimodal content at varying granularities. To operationalize these strategies, UniNote employs a two-stage training paradigm: the first stage leverages contrastive SFT to establish robust base embeddings, while the second stage refines ranking quality through a reinforcement learning (RL) process that aligns the model with content relevance. Our results show that UniNote achieves SOTA performance across diverse I2I tasks. Deployed at Xiaohongshu and integrated with Matryoshka Representation Learning (MRL), UniNote achieved significant improvements in retrieval quality and cost efficiency in large-scale applications.

Jintao Tong, Jiaqi Gu, Yujing Lou et al.

While Multimodal Large Language Models (MLLMs) excel at visual understanding tasks through text reasoning, they often fall short in scenarios requiring visual imagination. Unlike current works that take predefined external toolkits or generate images during thinking, however, humans can form flexible visual-text imagination and interactions during thinking without predefined toolkits, where one important reason is that humans construct the visual-text thinking process in a unified space inside the brain. Inspired by this capability, given that current MLLMs already encode visual and text information in the same feature space, we hold that visual tokens can be seamlessly inserted into the reasoning process carried by text tokens, where ideally, all visual imagination processes can be encoded by the latent features. To achieve this goal, we propose Sketch-in-Latents (SkiLa), a novel paradigm for unified multi-modal reasoning that expands the auto-regressive capabilities of MLLMs to natively generate continuous visual embeddings, termed latent sketch tokens, as visual thoughts. During multi-step reasoning, the model dynamically alternates between textual thinking mode for generating textual think tokens and visual sketching mode for generating latent sketch tokens. A latent visual semantics reconstruction mechanism is proposed to ensure these latent sketch tokens are semantically grounded. Extensive experiments demonstrate that SkiLa achieves superior performance on vision-centric tasks while exhibiting strong generalization to diverse general multi-modal benchmarks. Codes will be released at https://github.com/TungChintao/SkiLa.

Jintao Tong, Shilin Yan, Hongwei Xue et al.

Multimodal Large Language Models (MLLMs) have made remarkable progress in multimodal perception and reasoning by bridging vision and language. However, most existing MLLMs perform reasoning primarily with textual CoT, which limits their effectiveness on vision-intensive tasks. Recent approaches inject a fixed number of continuous hidden states as "visual thoughts" into the reasoning process and improve visual performance, but often at the cost of degraded text-based logical reasoning. We argue that the core limitation lies in a rigid, pre-defined reasoning pattern that cannot adaptively choose the most suitable thinking modality for different user queries. We introduce SwimBird, a reasoning-switchable MLLM that dynamically switches among three reasoning modes conditioned on the input: (1) text-only reasoning, (2) vision-only reasoning (continuous hidden states as visual thoughts), and (3) interleaved vision-text reasoning. To enable this capability, we adopt a hybrid autoregressive formulation that unifies next-token prediction for textual thoughts with next-embedding prediction for visual thoughts, and design a systematic reasoning-mode curation strategy to construct SwimBird-SFT-92K, a diverse supervised fine-tuning dataset covering all three reasoning patterns. By enabling flexible, query-adaptive mode selection, SwimBird preserves strong textual logic while substantially improving performance on vision-dense tasks. Experiments across diverse benchmarks covering textual reasoning and challenging visual understanding demonstrate that SwimBird achieves state-of-the-art results and robust gains over prior fixed-pattern multimodal reasoning methods.

Anqi Li, Wenwei Jin, Jintao Tong et al.

Social platforms have revolutionized information sharing, but also accelerated the dissemination of harmful and policy-violating content. To ensure safety and compliance at scale, moderation systems must go beyond efficiency and offer accuracy and interpretability. However, current approaches largely rely on noisy, label-driven learning, lacking alignment with moderation rules and producing opaque decisions that hinder human review. Therefore, we propose Hierarchical Guard (Hi-Guard), a multimodal moderation framework that introduces a new policy-aligned decision paradigm. The term "Hierarchical" reflects two key aspects of our system design: (1) a hierarchical moderation pipeline, where a lightweight binary model first filters safe content and a stronger model handles fine-grained risk classification; and (2) a hierarchical taxonomy in the second stage, where the model performs path-based classification over a hierarchical taxonomy ranging from coarse to fine-grained levels. To ensure alignment with evolving moderation policies, Hi-Guard directly incorporates rule definitions into the model prompt. To further enhance structured prediction and reasoning, we introduce a multi-level soft-margin reward and optimize with Group Relative Policy Optimization (GRPO), penalizing semantically adjacent misclassifications and improving explanation quality. Extensive experiments and real-world deployment demonstrate that Hi-Guard achieves superior classification accuracy, generalization, and interpretability, paving the way toward scalable, transparent, and trustworthy content safety systems. Code is available at: https://github.com/lianqi1008/Hi-Guard.

Jintao Tong, Wenwei Jin, Pengda Qin et al.

Large vision-language models (LVLMs) excel at multimodal understanding but suffer from high computational costs due to redundant vision tokens. Existing pruning methods typically rely on single-layer attention scores to rank and prune redundant visual tokens to solve this inefficiency. However, as the interaction between tokens and layers is complicated, this raises a basic question: Is such a simple single-layer criterion sufficient to identify redundancy? To answer this question, we rethink the emergence of redundant visual tokens from a fundamental perspective: information flow, which models the interaction between tokens and layers by capturing how information moves between tokens across layers. We find (1) the CLS token acts as an information relay, which can simplify the complicated flow analysis; (2) the redundancy emerges progressively and dynamically via layer-wise attention concentration; and (3) relying solely on attention scores from single layers can lead to contradictory redundancy identification. Based on this, we propose FlowCut, an information-flow-aware pruning framework, mitigating the insufficiency of the current criterion for identifying redundant tokens and better aligning with the model's inherent behaviors. Extensive experiments show that FlowCut achieves superior results, outperforming SoTA by 1.6% on LLaVA-1.5-7B with 88.9% token reduction, and by 4.3% on LLaVA-NeXT-7B with 94.4% reduction, delivering 3.2x speed-up in the prefilling stage. Our code is available at https://github.com/TungChintao/FlowCut

Shilin Yan, Jintao Tong, Hongwei Xue et al.

The advent of agentic multimodal models has empowered systems to actively interact with external environments. However, current agents suffer from a profound meta-cognitive deficit: they struggle to arbitrate between leveraging internal knowledge and querying external utilities. Consequently, they frequently fall prey to blind tool invocation, resorting to reflexive tool execution even when queries are resolvable from the raw visual context. This pathological behavior precipitates severe latency bottlenecks and injects extraneous noise that derails sound reasoning. Existing reinforcement learning protocols attempt to mitigate this via a scalarized reward that penalizes tool usage. Yet, this coupled formulation creates an irreconcilable optimization dilemma: an aggressive penalty suppresses essential tool use, whereas a mild penalty is entirely subsumed by the variance of the accuracy reward during advantage normalization, rendering it impotent against tool overuse. To transcend this bottleneck, we propose HDPO, a framework that reframes tool efficiency from a competing scalar objective to a strictly conditional one. By eschewing reward scalarization, HDPO maintains two orthogonal optimization channels: an accuracy channel that maximizes task correctness, and an efficiency channel that enforces execution economy exclusively within accurate trajectories via conditional advantage estimation. This decoupled architecture naturally induces a cognitive curriculum-compelling the agent to first master task resolution before refining its self-reliance. Extensive evaluations demonstrate that our resulting model, Metis, reduces tool invocations by orders of magnitude while simultaneously elevating reasoning accuracy.

Jintao Tong, Yixiong Zou, Yuhua Li et al.

Cross-domain few-shot segmentation (CD-FSS) is proposed to first pre-train the model on a large-scale source-domain dataset, and then transfer the model to data-scarce target-domain datasets for pixel-level segmentation. The significant domain gap between the source and target datasets leads to a sharp decline in the performance of existing few-shot segmentation (FSS) methods in cross-domain scenarios. In this work, we discover an intriguing phenomenon: simply filtering different frequency components for target domains can lead to a significant performance improvement, sometimes even as high as 14% mIoU. Then, we delve into this phenomenon for an interpretation, and find such improvements stem from the reduced inter-channel correlation in feature maps, which benefits CD-FSS with enhanced robustness against domain gaps and larger activated regions for segmentation. Based on this, we propose a lightweight frequency masker, which further reduces channel correlations by an Amplitude-Phase Masker (APM) module and an Adaptive Channel Phase Attention (ACPA) module. Notably, APM introduces only 0.01% additional parameters but improves the average performance by over 10%, and ACPA imports only 2.5% parameters but further improves the performance by over 1.5%, which significantly surpasses the state-of-the-art CD-FSS methods.

Jintao Tong, Ran Ma, Yixiong Zou et al.

Cross-domain few-shot segmentation (CD-FSS) is proposed to pre-train the model on a source-domain dataset with sufficient samples, and then transfer the model to target-domain datasets where only a few samples are available for efficient fine-tuning. There are majorly two challenges in this task: (1) the domain gap and (2) fine-tuning with scarce data. To solve these challenges, we revisit the adapter-based methods, and discover an intriguing insight not explored in previous works: the adapter not only helps the fine-tuning of downstream tasks but also naturally serves as a domain information decoupler. Then, we delve into this finding for an interpretation, and find the model's inherent structure could lead to a natural decoupling of domain information. Building upon this insight, we propose the Domain Feature Navigator (DFN), which is a structure-based decoupler instead of loss-based ones like current works, to capture domain-specific information, thereby directing the model's attention towards domain-agnostic knowledge. Moreover, to prevent the potential excessive overfitting of DFN during the source-domain training, we further design the SAM-SVN method to constrain DFN from learning sample-specific knowledge. On target domains, we freeze the model and fine-tune the DFN to learn target-specific knowledge specific. Extensive experiments demonstrate that our method surpasses the state-of-the-art method in CD-FSS significantly by 2.69% and 4.68% MIoU in 1-shot and 5-shot scenarios, respectively.

Jintao Tong, Yixiong Zou, Guangyao Chen et al.

Cross-Domain Few-Shot Segmentation (CD-FSS) aims to transfer knowledge from a source-domain dataset to unseen target-domain datasets with limited annotations. Current methods typically compare the distance between training and testing samples for mask prediction. However, we find an entanglement problem exists in this widely adopted method, which tends to bind sourcedomain patterns together and make each of them hard to transfer. In this paper, we aim to address this problem for the CD-FSS task. We first find a natural decomposition of the ViT structure, based on which we delve into the entanglement problem for an interpretation. We find the decomposed ViT components are crossly compared between images in distance calculation, where the rational comparisons are entangled with those meaningless ones by their equal importance, leading to the entanglement problem. Based on this interpretation, we further propose to address the entanglement problem by learning to weigh for all comparisons of ViT components, which learn disentangled features and re-compose them for the CD-FSS task, benefiting both the generalization and finetuning. Experiments show that our model outperforms the state-of-the-art CD-FSS method by 1.92% and 1.88% in average accuracy under 1-shot and 5-shot settings, respectively.