Yuhang Hu

Robert Kwiatkowski, Yuhang Hu, Boyuan Chen et al.

Self-Modeling is the process by which an agent, such as an animal or machine, learns to create a predictive model of its own dynamics. Once captured, this self-model can then allow the agent to plan and evaluate various potential behaviors internally using the self-model, rather than using costly physical experimentation. Here, we quantify the benefits of such self-modeling against the complexity of the robot. We find a R2 =0.90 correlation between the number of degrees of freedom a robot has, and the added value of self-modeling as compared to a direct learning baseline. This result may help motivate self modeling in increasingly complex robotic systems, as well as shed light on the origins of self-modeling, and ultimately self-awareness, in animals and humans.

Zhenyu Yang, Kairui Zhang, Yuhang Hu et al.

Despite significant progress in Video Large Language Models (Video-LLMs) for offline video understanding, existing online Video-LLMs typically struggle to simultaneously process continuous frame-by-frame inputs and determine optimal response timing, often compromising real-time responsiveness and narrative coherence. To address these limitations, we introduce LiveStar, a pioneering live streaming assistant that achieves always-on proactive responses through adaptive streaming decoding. Specifically, LiveStar incorporates: (1) a training strategy enabling incremental video-language alignment for variable-length video streams, preserving temporal consistency across dynamically evolving frame sequences; (2) a response-silence decoding framework that determines optimal proactive response timing via a single forward pass verification; (3) memory-aware acceleration via peak-end memory compression for online inference on 10+ minute videos, combined with streaming key-value cache to achieve 1.53x faster inference. We also construct an OmniStar dataset, a comprehensive dataset for training and benchmarking that encompasses 15 diverse real-world scenarios and 5 evaluation tasks for online video understanding. Extensive experiments across three benchmarks demonstrate LiveStar's state-of-the-art performance, achieving an average 19.5% improvement in semantic correctness with 18.1% reduced timing difference compared to existing online Video-LLMs, while improving FPS by 12.0% across all five OmniStar tasks. Our model and dataset can be accessed at https://github.com/yzy-bupt/LiveStar.

Xin Li, Kun Yuan, Yajing Pei et al.

This paper reviews the NTIRE 2024 Challenge on Shortform UGC Video Quality Assessment (S-UGC VQA), where various excellent solutions are submitted and evaluated on the collected dataset KVQ from popular short-form video platform, i.e., Kuaishou/Kwai Platform. The KVQ database is divided into three parts, including 2926 videos for training, 420 videos for validation, and 854 videos for testing. The purpose is to build new benchmarks and advance the development of S-UGC VQA. The competition had 200 participants and 13 teams submitted valid solutions for the final testing phase. The proposed solutions achieved state-of-the-art performances for S-UGC VQA. The project can be found at https://github.com/lixinustc/KVQChallenge-CVPR-NTIRE2024.

Yuhang Hu, Jiong Lin, Hod Lipson

The emergence of vision catalysed a pivotal evolutionary advancement, enabling organisms not only to perceive but also to interact intelligently with their environment. This transformation is mirrored by the evolution of robotic systems, where the ability to leverage vision to simulate and predict their own dynamics marks a leap towards autonomy and self-awareness. Humans utilize vision to record experiences and internally simulate potential actions. For example, we can imagine that, if we stand up and raise our arms, the body will form a T shape without physical movement. Similarly, simulation allows robots to plan and predict the outcomes of potential actions without execution. Here we introduce a self-supervised learning framework to enable robots to model and predict their morphology, kinematics and motor control using only brief raw video data, eliminating the need for extensive real-world data collection and kinematic priors. By observing their own movements, akin to humans watching their reflection in a mirror, robots learn an ability to simulate themselves and predict their spatial motion for various tasks. Our results demonstrate that this self-learned simulation not only enables accurate motion planning but also allows the robot to detect abnormalities and recover from damage.

Zhenyu Yang, Yuhang Hu, Zemin Du et al.

Despite the significant advancements of Large Vision-Language Models (LVLMs) on established benchmarks, there remains a notable gap in suitable evaluation regarding their applicability in the emerging domain of long-context streaming video understanding. Current benchmarks for video understanding typically emphasize isolated single-instance text inputs and fail to evaluate the capacity to sustain temporal reasoning throughout the entire duration of video streams. To address these limitations, we introduce SVBench, a pioneering benchmark with temporal multi-turn question-answering chains specifically designed to thoroughly assess the capabilities of streaming video understanding of current LVLMs. We design a semi-automated annotation pipeline to obtain 49,979 Question-Answer (QA) pairs of 1,353 streaming videos, which includes generating QA chains that represent a series of consecutive multi-turn dialogues over video segments and constructing temporal linkages between successive QA chains. Our experimental results, obtained from 14 models in dialogue and streaming evaluations, reveal that while the closed-source GPT-4o outperforms others, most open-source LVLMs struggle with long-context streaming video understanding. We also construct a StreamingChat model, which significantly outperforms open-source LVLMs on our SVBench and achieves comparable performance on diverse vision-language benchmarks. We expect SVBench to advance the research of streaming video understanding by providing a comprehensive and in-depth analysis of current LVLMs. Our benchmark and model can be accessed at https://github.com/sotayang/SVBench.

Judah Goldfeder, Shreyes Kaliyur, Vaibhav Sourirajan et al.

Data augmentation has long been a cornerstone for reducing overfitting in vision models, with methods like AutoAugment automating the design of task-specific augmentations. Recent advances in generative models, such as conditional diffusion and few-shot NeRFs, offer a new paradigm for data augmentation by synthesizing data with significantly greater diversity and realism. However, unlike traditional augmentations like cropping or rotation, these methods introduce substantial changes that enhance robustness but also risk degrading performance if the augmentations are poorly matched to the task. In this work, we present EvoAug, an automated augmentation learning pipeline, which leverages these generative models alongside an efficient evolutionary algorithm to learn optimal task-specific augmentations. Our pipeline introduces a novel approach to image augmentation that learns stochastic augmentation trees that hierarchically compose augmentations, enabling more structured and adaptive transformations. We demonstrate strong performance across fine-grained classification and few-shot learning tasks. Notably, our pipeline discovers augmentations that align with domain knowledge, even in low-data settings. These results highlight the potential of learned generative augmentations, unlocking new possibilities for robust model training.

Yuhang Hu, Judah Goldfeder, Zhizhuo Zhang et al.

For robots to truly collaborate and assist humans, they must understand not only logic and instructions, but also the subtle emotions, aesthetics, and feelings that define our humanity. Human art and aesthetics are among the most elusive concepts-often difficult even for people to articulate-and without grasping these fundamentals, robots will be unable to help in many spheres of daily life. Consider the long-promised robotic butler: automating domestic chores demands more than motion planning. It requires an internal model of cleanliness and tidiness-a challenge largely unexplored by AI. To bridge this gap, we propose an approach that equips domestic robots to perform simple tidying tasks via knolling, the practice of arranging scattered items into neat, space-efficient layouts. Unlike the uniformity of industrial settings, household environments feature diverse objects and highly subjective notions of tidiness. Drawing inspiration from NLP, we treat knolling as a sequential prediction problem and employ a transformer based model to forecast each object's placement. Our method learns a generalizable concept of tidiness, generates diverse solutions adaptable to varying object sets, and incorporates human preferences for personalized arrangements. This work represents a step forward in building robots that internalize human aesthetic sense and can genuinely co-create in our living spaces.

Hongyang Wei, Bin Wen, Yancheng Long et al.

We present UniRef-Image-Edit, a high-performance multi-modal generation system that unifies single-image editing and multi-image composition within a single framework. Existing diffusion-based editing methods often struggle to maintain consistency across multiple conditions due to limited interaction between reference inputs. To address this, we introduce Sequence-Extended Latent Fusion (SELF), a unified input representation that dynamically serializes multiple reference images into a coherent latent sequence. During a dedicated training stage, all reference images are jointly constrained to fit within a fixed-length sequence under a global pixel-budget constraint. Building upon SELF, we propose a two-stage training framework comprising supervised fine-tuning (SFT) and reinforcement learning (RL). In the SFT stage, we jointly train on single-image editing and multi-image composition tasks to establish a robust generative prior. We adopt a progressive sequence length training strategy, in which all input images are initially resized to a total pixel budget of $1024^2$, and are then gradually increased to $1536^2$ and $2048^2$ to improve visual fidelity and cross-reference consistency. This gradual relaxation of compression enables the model to incrementally capture finer visual details while maintaining stable alignment across references. For the RL stage, we introduce Multi-Source GRPO (MSGRPO), to our knowledge the first reinforcement learning framework tailored for multi-reference image generation. MSGRPO optimizes the model to reconcile conflicting visual constraints, significantly enhancing compositional consistency. We will open-source the code, models, training data, and reward data for community research purposes.

Yuhang Hu, Zhenyu Yang, Shihan Wang et al.

The rapid growth of streaming video applications demands multimodal models with enhanced capabilities for temporal dynamics understanding and complex reasoning. However, current Video Question Answering (VideoQA) datasets suffer from two critical limitations: 1) Static annotation mechanisms fail to capture the evolving nature of answers in temporal video streams, and 2) The absence of explicit reasoning process annotations restricts model interpretability and logical deduction capabilities. To address these challenges, We introduce StreamingCoT, the first dataset explicitly designed for temporally evolving reasoning in streaming VideoQA and multimodal Chain-of-Thought (CoT) tasks. Our framework first establishes a dynamic hierarchical annotation architecture that generates per-second dense descriptions and constructs temporally-dependent semantic segments through similarity fusion, paired with question-answer sets constrained by temporal evolution patterns. We further propose an explicit reasoning chain generation paradigm that extracts spatiotemporal objects via keyframe semantic alignment, derives object state transition-based reasoning paths using large language models, and ensures logical coherence through human-verified validation. This dataset establishes a foundation for advancing research in streaming video understanding, complex temporal reasoning, and multimodal inference. Our StreamingCoT and its construction toolkit can be accessed at https://github.com/Fleeting-hyh/StreamingCoT.

Kwai Keye Team, Biao Yang, Bin Wen et al.

While Multimodal Large Language Models (MLLMs) demonstrate remarkable capabilities on static images, they often fall short in comprehending dynamic, information-dense short-form videos, a dominant medium in today's digital landscape. To bridge this gap, we introduce \textbf{Kwai Keye-VL}, an 8-billion-parameter multimodal foundation model engineered for leading-edge performance in short-video understanding while maintaining robust general-purpose vision-language abilities. The development of Keye-VL rests on two core pillars: a massive, high-quality dataset exceeding 600 billion tokens with a strong emphasis on video, and an innovative training recipe. This recipe features a four-stage pre-training process for solid vision-language alignment, followed by a meticulous two-phase post-training process. The first post-training stage enhances foundational capabilities like instruction following, while the second phase focuses on stimulating advanced reasoning. In this second phase, a key innovation is our five-mode ``cold-start'' data mixture, which includes ``thinking'', ``non-thinking'', ``auto-think'', ``think with image'', and high-quality video data. This mixture teaches the model to decide when and how to reason. Subsequent reinforcement learning (RL) and alignment steps further enhance these reasoning capabilities and correct abnormal model behaviors, such as repetitive outputs. To validate our approach, we conduct extensive evaluations, showing that Keye-VL achieves state-of-the-art results on public video benchmarks and remains highly competitive on general image-based tasks (Figure 1). Furthermore, we develop and release the \textbf{KC-MMBench}, a new benchmark tailored for real-world short-video scenarios, where Keye-VL shows a significant advantage.

Biao Yang, Bin Wen, Boyang Ding et al.

In recent years, the development of Large Language Models (LLMs) has significantly advanced, extending their capabilities to multimodal tasks through Multimodal Large Language Models (MLLMs). However, video understanding remains a challenging area due to the dynamic and information-dense nature of videos. Existing models struggle with the trade-off between spatial resolution and temporal coverage when processing video content. We present Keye-VL-1.5, which addresses fundamental challenges in video comprehension through three key innovations. First, we introduce a novel Slow-Fast video encoding strategy that dynamically allocates computational resources based on inter-frame similarity, processing key frames with significant visual changes at higher resolution (Slow pathway) while handling relatively static frames with increased temporal coverage at lower resolution (Fast pathway). Second, we implement a progressive four-stage pre-training methodology that systematically extends the model's context length from 8K to 128K tokens, enabling processing of longer videos and more complex visual content. Third, we develop a comprehensive post-training pipeline focusing on reasoning enhancement and human preference alignment, incorporating a 5-step chain-of-thought data construction process, iterative GSPO-based reinforcement learning with progressive prompt hinting for difficult cases, and alignment training. Through extensive evaluation on public benchmarks and rigorous internal human assessment, Keye-VL-1.5 demonstrates significant improvements over existing models, particularly excelling in video understanding tasks while maintaining competitive performance on general multimodal benchmarks.

Xingyu Lu, Yuhang Hu, Changyi Liu et al.

Mathematical reasoning presents a significant challenge to the cognitive capabilities of LLMs. Various methods have been proposed to enhance the mathematical ability of LLMs. However, few recognize the value of state transition for LLM reasoning. In this work, we define mathematical problem-solving as a process of transiting from an initial unsolved state to the final resolved state, and propose Kwai-STaR framework, which transforms LLMs into State-Transition Reasoners to improve their intuitive reasoning capabilities. Our approach comprises three main steps: (1) Define the state space tailored to the mathematical reasoning. (2) Generate state-transition data based on the state space. (3) Convert original LLMs into State-Transition Reasoners via a curricular training strategy. Our experiments validate the effectiveness of Kwai-STaR in enhancing mathematical reasoning: After training on the small-scale Kwai-STaR dataset, general LLMs, including Mistral-7B and LLaMA-3, achieve considerable performance gain on the GSM8K and GSM-Hard dataset. Additionally, the state transition-based design endows Kwai-STaR with remarkable training and inference efficiency. Further experiments are underway to establish the generality of Kwai-STaR.

Xingyu Lu, Yuhang Hu, YiFan Zhang et al.

Reinforcement Learning from Human/AI Feedback (RLHF/RLAIF) has been extensively utilized for preference alignment of text-to-image models. Existing methods face certain limitations in terms of both data and algorithm. For training data, most approaches rely on manual annotated preference data, either by directly fine-tuning the generators or by training reward models to provide training signals. However, the high annotation cost makes them difficult to scale up, the reward model consumes extra computation and cannot guarantee accuracy. From an algorithmic perspective, most methods neglect the value of text and only take the image feedback as a comparative signal, which is inefficient and sparse. To alleviate these drawbacks, we propose the InstructEngine framework. Regarding annotation cost, we first construct a taxonomy for text-to-image generation, then develop an automated data construction pipeline based on it. Leveraging advanced large multimodal models and human-defined rules, we generate 25K text-image preference pairs. Finally, we introduce cross-validation alignment method, which refines data efficiency by organizing semantically analogous samples into mutually comparable pairs. Evaluations on DrawBench demonstrate that InstructEngine improves SD v1.5 and SDXL's performance by 10.53% and 5.30%, outperforming state-of-the-art baselines, with ablation study confirming the benefits of InstructEngine's all components. A win rate of over 50% in human reviews also proves that InstructEngine better aligns with human preferences.

Boyuan Chen, Yuhang Hu, Lianfeng Li et al.

Ability to generate intelligent and generalizable facial expressions is essential for building human-like social robots. At present, progress in this field is hindered by the fact that each facial expression needs to be programmed by humans. In order to adapt robot behavior in real time to different situations that arise when interacting with human subjects, robots need to be able to train themselves without requiring human labels, as well as make fast action decisions and generalize the acquired knowledge to diverse and new contexts. We addressed this challenge by designing a physical animatronic robotic face with soft skin and by developing a vision-based self-supervised learning framework for facial mimicry. Our algorithm does not require any knowledge of the robot's kinematic model, camera calibration or predefined expression set. By decomposing the learning process into a generative model and an inverse model, our framework can be trained using a single motor babbling dataset. Comprehensive evaluations show that our method enables accurate and diverse face mimicry across diverse human subjects. The project website is at http://www.cs.columbia.edu/~bchen/aiface/

Boyuan Chen, Yuhang Hu, Robert Kwiatkowski et al.

In order to engage in complex social interaction, humans learn at a young age to infer what others see and cannot see from a different point-of-view, and learn to predict others' plans and behaviors. These abilities have been mostly lacking in robots, sometimes making them appear awkward and socially inept. Here we propose an end-to-end long-term visual prediction framework for robots to begin to acquire both these critical cognitive skills, known as Visual Perspective Taking (VPT) and Theory of Behavior (TOB). We demonstrate our approach in the context of visual hide-and-seek - a game that represents a cognitive milestone in human development. Unlike traditional visual predictive model that generates new frames from immediate past frames, our agent can directly predict to multiple future timestamps (25s), extrapolating by 175% beyond the training horizon. We suggest that visual behavior modeling and perspective taking skills will play a critical role in the ability of physical robots to fully integrate into real-world multi-agent activities. Our website is at http://www.cs.columbia.edu/~bchen/vpttob/.