Yuxin Song

Bo Fang, Wenhao Wu, Chang Liu et al. · amazon-science

With the explosive growth of web videos and emerging large-scale vision-language pre-training models, e.g., CLIP, retrieving videos of interest with text instructions has attracted increasing attention. A common practice is to transfer text-video pairs to the same embedding space and craft cross-modal interactions with certain entities in specific granularities for semantic correspondence. Unfortunately, the intrinsic uncertainties of optimal entity combinations in appropriate granularities for cross-modal queries are understudied, which is especially critical for modalities with hierarchical semantics, e.g., video, text, etc. In this paper, we propose an Uncertainty-Adaptive Text-Video Retrieval approach, termed UATVR, which models each look-up as a distribution matching procedure. Concretely, we add additional learnable tokens in the encoders to adaptively aggregate multi-grained semantics for flexible high-level reasoning. In the refined embedding space, we represent text-video pairs as probabilistic distributions where prototypes are sampled for matching evaluation. Comprehensive experiments on four benchmarks justify the superiority of our UATVR, which achieves new state-of-the-art results on MSR-VTT (50.8%), VATEX (64.5%), MSVD (49.7%), and DiDeMo (45.8%). The code is available at https://github.com/bofang98/UATVR.

Wenhao Wu, Yuxin Song, Zhun Sun et al. · amazon-science

Temporal modeling plays a crucial role in understanding video content. To tackle this problem, previous studies built complicated temporal relations through time sequence thanks to the development of computationally powerful devices. In this work, we explore the potential of four simple arithmetic operations for temporal modeling. Specifically, we first capture auxiliary temporal cues by computing addition, subtraction, multiplication, and division between pairs of extracted frame features. Then, we extract corresponding features from these cues to benefit the original temporal-irrespective domain. We term such a simple pipeline as an Arithmetic Temporal Module (ATM), which operates on the stem of a visual backbone with a plug-and-play style. We conduct comprehensive ablation studies on the instantiation of ATMs and demonstrate that this module provides powerful temporal modeling capability at a low computational cost. Moreover, the ATM is compatible with both CNNs- and ViTs-based architectures. Our results show that ATM achieves superior performance over several popular video benchmarks. Specifically, on Something-Something V1, V2 and Kinetics-400, we reach top-1 accuracy of 65.6%, 74.6%, and 89.4% respectively. The code is available at https://github.com/whwu95/ATM.

Wenhao Wu, Huanjin Yao, Mengxi Zhang et al. · amazon-science

This paper does not present a novel method. Instead, it delves into an essential, yet must-know baseline in light of the latest advancements in Generative Artificial Intelligence (GenAI): the utilization of GPT-4 for visual understanding. Our study centers on the evaluation of GPT-4's linguistic and visual capabilities in zero-shot visual recognition tasks: Firstly, we explore the potential of its generated rich textual descriptions across various categories to enhance recognition performance without any training. Secondly, we evaluate GPT-4's visual proficiency in directly recognizing diverse visual content. We conducted extensive experiments to systematically evaluate GPT-4's performance across images, videos, and point clouds, using 16 benchmark datasets to measure top-1 and top-5 accuracy. Our findings show that GPT-4, enhanced with rich linguistic descriptions, significantly improves zero-shot recognition, offering an average top-1 accuracy increase of 7% across all datasets. GPT-4 excels in visual recognition, outshining OpenAI-CLIP's ViT-L and rivaling EVA-CLIP's ViT-E, particularly in video datasets HMDB-51 and UCF-101, where it leads by 22% and 9%, respectively. We hope this research contributes valuable data points and experience for future studies. We release our code at https://github.com/whwu95/GPT4Vis.

ByteDance Seed, Jiaze Chen, Tiantian Fan et al. · bytedance

We introduce Seed1.5-Thinking, capable of reasoning through thinking before responding, resulting in improved performance on a wide range of benchmarks. Seed1.5-Thinking achieves 86.7 on AIME 2024, 55.0 on Codeforces and 77.3 on GPQA, demonstrating excellent reasoning abilities in STEM and coding. Beyond reasoning tasks, the method demonstrates notable generalization across diverse domains. For instance, it surpasses DeepSeek R1 by 8% in win rate on non-reasoning tasks, indicating its broader applicability. Compared to other state-of-the-art reasoning models, Seed1.5-Thinking is a Mixture-of-Experts (MoE) model with a relatively small size, featuring 20B activated and 200B total parameters. As part of our effort to assess generalized reasoning, we develop two internal benchmarks, BeyondAIME and Codeforces, both of which will be publicly released to support future research. Model trial link: https://www.volcengine.com/experience/ark.

Siyang Song, Yuxin Song, Cheng Luo et al.

Graph is powerful for representing various types of real-world data. The topology (edges' presence) and edges' features of a graph decides the message passing mechanism among vertices within the graph. While most existing approaches only manually define a single-value edge to describe the connectivity or strength of association between a pair of vertices, task-specific and crucial relationship cues may be disregarded by such manually defined topology and single-value edge features. In this paper, we propose the first general graph representation learning framework (called GRATIS) which can generate a strong graph representation with a task-specific topology and task-specific multi-dimensional edge features from any arbitrary input. To learn each edge's presence and multi-dimensional feature, our framework takes both of the corresponding vertices pair and their global contextual information into consideration, enabling the generated graph representation to have a globally optimal message passing mechanism for different down-stream tasks. The principled investigation results achieved for various graph analysis tasks on 11 graph and non-graph datasets show that our GRATIS can not only largely enhance pre-defined graphs but also learns a strong graph representation for non-graph data, with clear performance improvements on all tasks. In particular, the learned topology and multi-dimensional edge features provide complementary task-related cues for graph analysis tasks. Our framework is effective, robust and flexible, and is a plug-and-play module that can be combined with different backbones and Graph Neural Networks (GNNs) to generate a task-specific graph representation from various graph and non-graph data. Our code is made publicly available at https://github.com/SSYSteve/Learning-Graph-Representation-with-Task-specific-Topology-and-Multi-dimensional-Edge-Features.

Yuxin Song, Min Yang, Wenhao Wu et al. · amazon-science

Self-supervised video transformer pre-training has recently benefited from the mask-and-predict pipeline. They have demonstrated outstanding effectiveness on downstream video tasks and superior data efficiency on small datasets. However, temporal relation is not fully exploited by these methods. In this work, we explicitly investigate motion cues in videos as extra prediction target and propose our Masked Appearance-Motion Modeling (MAM2) framework. Specifically, we design an encoder-regressor-decoder pipeline for this task. The regressor separates feature encoding and pretext tasks completion, such that the feature extraction process is completed adequately by the encoder. In order to guide the encoder to fully excavate spatial-temporal features, two separate decoders are used for two pretext tasks of disentangled appearance and motion prediction. We explore various motion prediction targets and figure out RGB-difference is simple yet effective. As for appearance prediction, VQGAN codes are leveraged as prediction target. With our pre-training pipeline, convergence can be remarkably speed up, e.g., we only require half of epochs than state-of-the-art VideoMAE (400 v.s. 800) to achieve the competitive performance. Extensive experimental results prove that our method learns generalized video representations. Notably, our MAM2 with ViT-B achieves 82.3% on Kinects-400, 71.3% on Something-Something V2, 91.5% on UCF101, and 62.5% on HMDB51.

Xinyao Zhang, Wenkai Dong, Yuxin Song et al.

Current instruction-guided video editing models struggle to simultaneously balance precise semantic modifications with faithful motion preservation. While existing approaches rely on injecting explicit external priors (e.g., VLM features or structural conditions) to mitigate these issues, this reliance severely bottlenecks model robustness and generalization. To overcome this limitation, we present SAMA (factorized Semantic Anchoring and Motion Alignment), a framework that factorizes video editing into semantic anchoring and motion modeling. First, we introduce Semantic Anchoring, which establishes a reliable visual anchor by jointly predicting semantic tokens and video latents at sparse anchor frames, enabling purely instruction-aware structural planning. Second, Motion Alignment pre-trains the same backbone on motion-centric video restoration pretext tasks (cube inpainting, speed perturbation, and tube shuffle), enabling the model to internalize temporal dynamics directly from raw videos. SAMA is optimized with a two-stage pipeline: a factorized pre-training stage that learns inherent semantic-motion representations without paired video-instruction editing data, followed by supervised fine-tuning on paired editing data. Remarkably, the factorized pre-training alone already yields strong zero-shot video editing ability, validating the proposed factorization. SAMA achieves state-of-the-art performance among open-source models and is competitive with leading commercial systems (e.g., Kling-Omni). Code, models, and datasets will be released.

Yuxin Song, Ruolin Zhu, Min Yang et al.

Deeply learned representations have achieved superior image retrieval performance in a retrieve-then-rerank manner. Recent state-of-the-art single stage model, which heuristically fuses local and global features, achieves promising trade-off between efficiency and effectiveness. However, we notice that efficiency of existing solutions is still restricted because of their multi-scale inference paradigm. In this paper, we follow the single stage art and obtain further complexity-effectiveness balance by successfully getting rid of multi-scale testing. To achieve this goal, we abandon the widely-used convolution network giving its limitation in exploring diverse visual patterns, and resort to fully attention based framework for robust representation learning motivated by the success of Transformer. Besides applying Transformer for global feature extraction, we devise a local branch composed of window-based multi-head attention and spatial attention to fully exploit local image patterns. Furthermore, we propose to combine the hierarchical local and global features via a cross-attention module, instead of using heuristically fusion as previous art does. With our Deep Attentive Local and Global modeling framework (DALG), extensive experimental results show that efficiency can be significantly improved while maintaining competitive results with the state of the arts.

Lei Wang, YuXin Song, Ge Wu et al.

Reference-to-video (R2V) generation is a controllable video synthesis paradigm that constrains the generation process using both text prompts and reference images, enabling applications such as personalized advertising and virtual try-on. In practice, existing R2V methods typically introduce additional high-level semantic or cross-modal features alongside the VAE latent representation of the reference image and jointly feed them into the diffusion Transformer (DiT). These auxiliary representations provide semantic guidance and act as implicit alignment signals, which can partially alleviate pixel-level information leakage in the VAE latent space. However, they may still struggle to address copy--paste artifacts and multi-subject confusion caused by modality mismatch across heterogeneous encoder features. In this paper, we propose RefAlign, a representation alignment framework that explicitly aligns DiT reference-branch features to the semantic space of a visual foundation model (VFM). The core of RefAlign is a reference alignment loss that pulls the reference features and VFM features of the same subject closer to improve identity consistency, while pushing apart the corresponding features of different subjects to enhance semantic discriminability. This simple yet effective strategy is applied only during training, incurring no inference-time overhead, and achieves a better balance between text controllability and reference fidelity. Extensive experiments on the OpenS2V-Eval benchmark demonstrate that RefAlign outperforms current state-of-the-art methods in TotalScore, validating the effectiveness of explicit reference alignment for R2V tasks.

Huanjin Yao, Jiaxing Huang, Wenhao Wu et al.

In this work, we aim to develop an MLLM that understands and solves questions by learning to create each intermediate step of the reasoning involved till the final answer. To this end, we propose Collective Monte Carlo Tree Search (CoMCTS), a new learning-to-reason method for MLLMs, which introduces the concept of collective learning into ``tree search'' for effective and efficient reasoning-path searching and learning. The core idea of CoMCTS is to leverage collective knowledge from multiple models to collaboratively conjecture, search and identify effective reasoning paths toward correct answers via four iterative operations including Expansion, Simulation and Error Positioning, Backpropagation, and Selection. Using CoMCTS, we construct Mulberry-260k, a multimodal dataset with a tree of rich, explicit and well-defined reasoning nodes for each question. With Mulberry-260k, we perform collective SFT to train our model, Mulberry, a series of MLLMs with o1-like step-by-step Reasoning and Reflection capabilities. Extensive experiments demonstrate the superiority of our proposed methods on various benchmarks. Code will be available at https://github.com/HJYao00/Mulberry

Huanjin Yao, Wenhao Wu, Taojiannan Yang et al.

Do we fully leverage the potential of visual encoder in Multimodal Large Language Models (MLLMs)? The recent outstanding performance of MLLMs in multimodal understanding has garnered broad attention from both academia and industry. In the current MLLM rat race, the focus seems to be predominantly on the linguistic side. We witness the rise of larger and higher-quality instruction datasets, as well as the involvement of larger-sized LLMs. Yet, scant attention has been directed towards the visual signals utilized by MLLMs, often assumed to be the final high-level features extracted by a frozen visual encoder. In this paper, we introduce the Dense Connector - a simple, effective, and plug-and-play vision-language connector that significantly enhances existing MLLMs by leveraging multi-layer visual features, with minimal additional computational overhead. Building on this, we also propose the Efficient Dense Connector, which achieves performance comparable to LLaVA-v1.5 with only 25% of the visual tokens. Furthermore, our model, trained solely on images, showcases remarkable zero-shot capabilities in video understanding as well. Experimental results across various vision encoders, image resolutions, training dataset scales, varying sizes of LLMs (2.7B->70B), and diverse architectures of MLLMs (e.g., LLaVA-v1.5, LLaVA-NeXT and Mini-Gemini) validate the versatility and scalability of our approach, achieving state-of-the-art performance across 19 image and video benchmarks. We hope that this work will provide valuable experience and serve as a basic module for future MLLM development. Code is available at https://github.com/HJYao00/DenseConnector .

Mengxi Zhang, Wenhao Wu, Yu Lu et al.

Current multimodal Large Language Models (MLLMs) suffer from ``hallucination'', occasionally generating responses that are not grounded in the input images. To tackle this challenge, one promising path is to utilize reinforcement learning from human feedback (RLHF), which steers MLLMs towards learning superior responses while avoiding inferior ones. We rethink the common practice of using binary preferences (i.e., superior, inferior), and find that adopting multi-level preferences (e.g., superior, medium, inferior) is better for two benefits: 1) It narrows the gap between adjacent levels, thereby encouraging MLLMs to discern subtle differences. 2) It further integrates cross-level comparisons (beyond adjacent-level comparisons), thus providing a broader range of comparisons with hallucination examples. To verify our viewpoint, we present the Automated Multi-level Preference (AMP) framework for MLLMs. To facilitate this framework, we first develop an automated dataset generation pipeline that provides high-quality multi-level preference datasets without any human annotators. Furthermore, we design the Multi-level Direct Preference Optimization (MDPO) algorithm to robustly conduct complex multi-level preference learning. Additionally, we propose a new hallucination benchmark, MRHal-Bench. Extensive experiments across public hallucination and general benchmarks, as well as our MRHal-Bench, demonstrate the effectiveness of our proposed method. Code is available at https://github.com/takomc/amp.

YuXin Song, Yu Lu, Haoyuan Sun et al.

Unified conditional image generation remains difficult because different tasks depend on fundamentally different internal representations. Some require conceptual understanding for semantic synthesis, while others rely on localization cues for spatial precision. Forcing these heterogeneous tasks to share a single representation leads to concept-localization representational conflict. To address this issue, we propose CoLoGen, a unified diffusion framework that progressively learns and reconciles this concept-localization duality. CoLoGen uses a staged curriculum that first builds core conceptual and localization abilities, then adapts them to diverse visual conditions, and finally refines their synergy for complex instruction-driven tasks. Central to this process is the Progressive Representation Weaving (PRW) module, which dynamically routes features to specialized experts and stably integrates their outputs across stages. Experiments on editing, controllable generation, and customized generation show that CoLoGen achieves competitive or superior performance, offering a principled representational perspective for unified image generation.

Huanjin Yao, Jiaxing Huang, Yawen Qiu et al.

Reasoning plays a crucial role in advancing Multimodal Large Language Models (MLLMs) toward Artificial General Intelligence. However, existing MLLM benchmarks often fall short in precisely and comprehensively evaluating long-chain reasoning abilities from three key aspects: (1) lack of difficulty and diversity, (2) susceptibility to guessability and memorization, (3) inadequate assessment of intermediate reasoning steps. To fill this gap, we introduce MMReason, a new benchmark designed to precisely and comprehensively evaluate MLLM long-chain reasoning capability with diverse, open-ended, challenging questions. First, we curate challenging questions requiring multi-step reasoning from various fields (i.e., 6 disciplines) and multiple difficulty levels (i.e., from pre-university to university, and from foundational to competition tiers). Second, these questions are reformulated into an open-ended format and filtered using a multi-model voting technique to eliminate shortcut cases related to guessing and memorization, ensuring robust reasoning evaluations. Third, we annotate the questions with detailed step-by-step solutions, and design a reference-based ternary scoring mechanism to reliably assess intermediate reasoning steps. With MMReason, we benchmark popular leading MLLMs and provide an in-depth analysis of their reasoning capabilities. We hope MMReason will serve as a valuable resource for advancing MLLM reasoning research. Code will be available at https://github.com/HJYao00/MMReason.

Bo Fang, Yuxin Song, Qiangqiang Wu et al.

Complex video reasoning remains a significant challenge for Multimodal Large Language Models (MLLMs), as current R1-based methodologies often prioritize text-centric reasoning derived from text-based and image-based developments. In video tasks, such strategies frequently underutilize rich visual information, leading to potential shortcut learning and increased susceptibility to hallucination. To foster a more robust, visual-centric video understanding, we start by introducing a novel self-supervised reinforcement learning GRPO algorithm (Pretext-GRPO) within the standard R1 pipeline, in which positive rewards are assigned for correctly solving pretext tasks on transformed visual inputs, which makes the model to non-trivially process the visual information. Building on the effectiveness of Pretext-GRPO, we further propose the ViSS-R1 framework, which streamlines and integrates pretext-task-based self-supervised learning directly into the MLLM's R1 post-training paradigm. Instead of relying solely on sparse visual cues, our framework compels models to reason about transformed visual input by simultaneously processing both pretext questions (concerning transformations) and true user queries. This necessitates identifying the applied transformation and reconstructing the original video to formulate accurate final answers. Comprehensive evaluations on six widely-used video reasoning and understanding benchmarks demonstrate the effectiveness and superiority of our Pretext-GRPO and ViSS-R1 for complex video reasoning. Our codes and models will be publicly available.

Haoyuan Sun, Jing Wang, Yuxin Song et al.

Recently, post-training methods based on reinforcement learning, with a particular focus on Group Relative Policy Optimization (GRPO), have emerged as the robust paradigm for further advancement of text-to-image (T2I) models. However, these methods are often prone to reward hacking, wherein models exploit biases in imperfect reward functions rather than yielding genuine performance gains. In this work, we identify that normalization could lead to miscalibration and directly removing the prompt-level standard deviation term yields an optimal policy ascent direction that is linear in the advantage but still limits the separation of genuine signals from noise. To mitigate the above issues, we propose Super-Linear Advantage Shaping (SLAS) by revisiting the functional update from an information geometry perspective. By extending the Fisher-Rao information metric with advantage-dependent weighting, SLAS introduces a non-linear geometric structure that reshapes the local policy space. This design relaxes constraints along high-advantage directions to amplify informative updates, while tightening those in low-advantage regions to suppress illusory gradients. In addition, batch-level normalization is applied to stabilize training under varying reward scales. Extensive evaluations demonstrate that SLAS consistently surpasses the DanceGRPO baseline across multiple backbones and benchmarks. In particular, it yields faster training dynamics, improved out-of-domain performance on GenEval and UniGenBench++, and enhanced robustness to model scaling, while mitigating reward hacking and preserving semantic and compositional fidelity in generations.

Yuxin Song, Wenkai Dong, Shizun Wang et al.

Unified Multimodal Models (UMMs) have demonstrated remarkable performance in text-to-image generation (T2I) and editing (TI2I), whether instantiated as assembled unified frameworks which couple powerful vision-language model (VLM) with diffusion-based generator, or as naive Unified Multimodal Models with an early fusion of understanding and generation modalities. We contend that in current unified frameworks, the crucial capability of multimodal generative reasoning which encompasses instruction understanding, grounding, and image referring for identity preservation and faithful reconstruction, is intrinsically entangled with high-fidelity synthesis. In this work, we introduce Query-Kontext, a novel approach that bridges the VLM and diffusion model via a multimodal ``kontext'' composed of semantic cues and coarse-grained image conditions encoded from multimodal inputs. This design delegates the complex ability of multimodal generative reasoning to powerful VLM while reserving diffusion model's role for high-quality visual synthesis. To achieve this, we propose a three-stage progressive training strategy. First, we connect the VLM to a lightweight diffusion head via multimodal kontext tokens to unleash the VLM's generative reasoning ability. Second, we scale this head to a large, pre-trained diffusion model to enhance visual detail and realism. Finally, we introduce a low-level image encoder to improve image fidelity and perform instruction tuning on downstream tasks. Furthermore, we build a comprehensive data pipeline integrating real, synthetic, and open-source datasets, covering diverse multimodal reference-to-image scenarios, including image generation, instruction-driven editing, customized generation, and multi-subject composition. Experiments show that our approach matches strong unified baselines and even outperforms task-specific state-of-the-art methods in several cases.

Keming Wu, Zuhao Yang, Kaichen Zhang et al.

Recent visual generation models have made major progress in photorealism, typography, instruction following, and interactive editing, yet they still struggle with spatial reasoning, persistent state, long-horizon consistency, and causal understanding. We argue that the field should move beyond appearance synthesis toward intelligent visual generation: plausible visuals grounded in structure, dynamics, domain knowledge, and causal relations. To frame this shift, we introduce a five-level taxonomy: Atomic Generation, Conditional Generation, In-Context Generation, Agentic Generation, and World-Modeling Generation, progressing from passive renderers to interactive, agentic, world-aware generators. We analyze key technical drivers, including flow matching, unified understanding-and-generation models, improved visual representations, post-training, reward modeling, data curation, synthetic data distillation, and sampling acceleration. We further show that current evaluations often overestimate progress by emphasizing perceptual quality while missing structural, temporal, and causal failures. By combining benchmark review, in-the-wild stress tests, and expert-constrained case studies, this roadmap offers a capability-centered lens for understanding, evaluating, and advancing the next generation of intelligent visual generation systems.

Haoming Wang, Haoyang Zou, Huatong Song et al. · pku

The development of autonomous agents for graphical user interfaces (GUIs) presents major challenges in artificial intelligence. While recent advances in native agent models have shown promise by unifying perception, reasoning, action, and memory through end-to-end learning, open problems remain in data scalability, multi-turn reinforcement learning (RL), the limitations of GUI-only operation, and environment stability. In this technical report, we present UI-TARS-2, a native GUI-centered agent model that addresses these challenges through a systematic training methodology: a data flywheel for scalable data generation, a stabilized multi-turn RL framework, a hybrid GUI environment that integrates file systems and terminals, and a unified sandbox platform for large-scale rollouts. Empirical evaluation demonstrates that UI-TARS-2 achieves significant improvements over its predecessor UI-TARS-1.5. On GUI benchmarks, it reaches 88.2 on Online-Mind2Web, 47.5 on OSWorld, 50.6 on WindowsAgentArena, and 73.3 on AndroidWorld, outperforming strong baselines such as Claude and OpenAI agents. In game environments, it attains a mean normalized score of 59.8 across a 15-game suite-roughly 60% of human-level performance-and remains competitive with frontier proprietary models (e.g., OpenAI o3) on LMGame-Bench. Additionally, the model can generalize to long-horizon information-seeking tasks and software engineering benchmarks, highlighting its robustness across diverse agent tasks. Detailed analyses of training dynamics further provide insights into achieving stability and efficiency in large-scale agent RL. These results underscore UI-TARS-2's potential to advance the state of GUI agents and exhibit strong generalization to real-world interactive scenarios.

Bo Fang, Wenhao Wu, Qiangqiang Wu et al.

Audio Descriptions (ADs) aim to provide a narration of a movie in text form, describing non-dialogue-related narratives, such as characters, actions, or scene establishment. Automatic generation of ADs remains challenging due to: i) the domain gap between movie-AD data and existing data used to train vision-language models, and ii) the issue of contextual redundancy arising from highly similar neighboring visual clips in a long movie. In this work, we propose DistinctAD, a novel two-stage framework for generating ADs that emphasize distinctiveness to produce better narratives. To address the domain gap, we introduce a CLIP-AD adaptation strategy that does not require additional AD corpora, enabling more effective alignment between movie and AD modalities at both global and fine-grained levels. In Stage-II, DistinctAD incorporates two key innovations: (i) a Contextual Expectation-Maximization Attention (EMA) module that reduces redundancy by extracting common bases from consecutive video clips, and (ii) an explicit distinctive word prediction loss that filters out repeated words in the context, ensuring the prediction of unique terms specific to the current AD. Comprehensive evaluations on MAD-Eval, CMD-AD, and TV-AD benchmarks demonstrate the superiority of DistinctAD, with the model consistently outperforming baselines, particularly in Recall@k/N, highlighting its effectiveness in producing high-quality, distinctive ADs.

Bo Fang, Wenhao Wu, Qiangqiang Wu et al.

Employing Multimodal Large Language Models (MLLMs) for long video understanding remains a challenging problem due to the dilemma between the substantial number of video frames (i.e., visual tokens) versus the limited context length of language models. Traditional uniform sampling often leads to selection of irrelevant content, while post-training MLLMs on thousands of frames imposes a substantial computational burden. In this paper, we propose threading keyframes with narratives (Nar-KFC), a plug-and-play module to facilitate effective and efficient long video perception. Nar-KFC generally involves two collaborative steps. First, we formulate the keyframe selection process as an integer quadratic programming problem, jointly optimizing query-relevance and frame-diversity. To avoid its computational complexity, a customized greedy search strategy is designed as an efficient alternative. Second, to mitigate the temporal discontinuity caused by sparse keyframe sampling, we further introduce interleaved textual narratives generated from non-keyframes using off-the-shelf captioners. These narratives are inserted between keyframes based on their true temporal order, forming a coherent and compact representation. Nar-KFC thus serves as a temporal- and content-aware compression strategy that complements visual and textual modalities. Experimental results on multiple long-video benchmarks demonstrate that Nar-KFC significantly improves the performance of popular MLLMs. Code will be made publicly available.

Yan Jia, Yuxin Song, Zihou Liu et al.

The Consumer Internet of Things (CIoT), a notable segment within the IoT domain, involves the integration of IoT technology into consumer electronics and devices, such as smart homes and smart wearables. Compared to traditional IoT fields, CIoT differs notably in target users, product types, and design approaches. While offering convenience to users, it also raises new security and privacy concerns. Network traffic analysis, a widely used technique in the security community, has been extensively applied to investigate these concerns about CIoT. Compared to traditional network traffic analysis in fields like mobile apps and websites, CIoT introduces unique characteristics that pose new challenges and research opportunities. Researchers have made significant contributions in this area. To aid researchers in understanding the application of traffic analysis tools for assessing CIoT security and privacy risks, this survey reviews 310 publications on traffic analysis within the CIoT security and privacy domain from January 2018 to June 2024, focusing on three research questions. Our work: 1) outlines the CIoT traffic analysis process and highlights its differences from general network traffic analysis. 2) summarizes and classifies existing research into four categories according to its application objectives: device fingerprinting, user activity inference, malicious traffic detection, and measurement. 3) explores emerging challenges and potential future research directions based on each step of the CIoT traffic analysis process. This will provide new insights to the community and guide the industry towards safer product designs.

Chen Zheng, Yiyuan Ma, Yuan Yang et al.

The development of alignment and reasoning capabilities in large language models has seen remarkable progress through two paradigms: instruction tuning and reinforcement learning from human feedback (RLHF) alignment paradigm, and distillation-based reasoning fine-tuning paradigm. While both approaches prove effective independently, the third paradigm of applying RLHF to distillation-trained models presents significant challenges. Our investigation reveals two critical phenomena that emerge in this paradigm: Sequence Length Collapse, where language generation dramatically reduces during early RLHF training, and the Reward Hockey Stick Curve, featuring severe reward score drops followed by gradual recovery. These instabilities fundamentally compromise the model's alignment and reasoning capabilities. To address these challenges, we propose Balanced Actor Initialization (BAI), a two-stage weighted model merging approach. BAI first merges instruction-following and distillation-based reasoning fine-tuned models, then further combines this intermediate model with the pretrained model to preserve foundational knowledge. Through comprehensive experiments across diverse benchmarks and detailed analysis of training experiments, we demonstrate that BAI resolves Sequence Length Collapse, mitigates the Reward Hockey Stick Curve, and enables continuous sequence length improvement during training. Additionally, our analysis reveals that balanced merging ratios achieve optimal trade-offs between training stability and reasoning capability preservation. Our work provides the effective solution for stable training in this third paradigm, enabling more capable reasoning models that combine distillation efficiency with RLHF alignment.

Yang Shen, Xiu-Shen Wei, Yifan Sun et al.

Computer Vision (CV) has yet to fully achieve the zero-shot task generalization observed in Natural Language Processing (NLP), despite following many of the milestones established in NLP, such as large transformer models, extensive pre-training, and the auto-regression paradigm, among others. In this paper, we explore the idea that CV adopts discrete and terminological task definitions (\eg, ``image segmentation''), which may be a key barrier to zero-shot task generalization. Our hypothesis is that without truly understanding previously-seen tasks--due to these terminological definitions--deep models struggle to generalize to novel tasks. To verify this, we introduce Explanatory Instructions, which provide an intuitive way to define CV task objectives through detailed linguistic transformations from input images to outputs. We create a large-scale dataset comprising 12 million ``image input $\to$ explanatory instruction $\to$ output'' triplets, and train an auto-regressive-based vision-language model (AR-based VLM) that takes both images and explanatory instructions as input. By learning to follow these instructions, the AR-based VLM achieves instruction-level zero-shot capabilities for previously-seen tasks and demonstrates strong zero-shot generalization for unseen CV tasks. Code and dataset will be openly available on our GitHub repository.

Jianjun Lei, Yuxin Song, Bo Peng et al.

Sketch-based image retrieval (SBIR) is a challenging task due to the large cross-domain gap between sketches and natural images. How to align abstract sketches and natural images into a common high-level semantic space remains a key problem in SBIR. In this paper, we propose a novel semi-heterogeneous three-way joint embedding network (Semi3-Net), which integrates three branches (a sketch branch, a natural image branch, and an edgemap branch) to learn more discriminative cross-domain feature representations for the SBIR task. The key insight lies with how we cultivate the mutual and subtle relationships amongst the sketches, natural images, and edgemaps. A semi-heterogeneous feature mapping is designed to extract bottom features from each domain, where the sketch and edgemap branches are shared while the natural image branch is heterogeneous to the other branches. In addition, a joint semantic embedding is introduced to embed the features from different domains into a common high-level semantic space, where all of the three branches are shared. To further capture informative features common to both natural images and the corresponding edgemaps, a co-attention model is introduced to conduct common channel-wise feature recalibration between different domains. A hybrid-loss mechanism is designed to align the three branches, where an alignment loss and a sketch-edgemap contrastive loss are presented to encourage the network to learn invariant cross-domain representations. Experimental results on two widely used category-level datasets (Sketchy and TU-Berlin Extension) demonstrate that the proposed method outperforms state-of-the-art methods.