arXiv subjects

Qile Su

Publications and source records attributed to Qile Su.

11 recordsLinked to original sources

arXiv2026

Thinking Beyond Videos: Unifying Video Reasoning and Deep Research for Open-World Video Agents

Open-world video understanding often requires a model to locate sparse visual evidence and acquire external knowledge that is absent from the video and its parametric memory. While Thinking-with-Videos enables active temporal perception and Deep Research supports multi-step information seeking, the two capabilities are typically developed in isolation. We introduce VideoRover, a unified Video Deep Research framework that iteratively coordinates video cropping, multimodal search, and webpage browsing. Given a video-question pair, VideoRover uses each tool result to select the next action, so localized video clips guide external retrieval and retrieved evidence triggers further video inspection and verification. To develop this capability, we construct an automated data curation pipeline, producing 26K verified SFT trajectories and 3K challenging RL instances. We also introduce VideoRover-Bench, a benchmark stratified by video duration and research difficulty. Experiments on VideoDR and VideoRover-Bench show that our VideoRover-8B-RL achieves performance comparable to proprietary models in the direct-answer setting without tool use while outperforming larger open-source models equipped with the same tool suite. Ablation studies and training dynamics further validate the complementary roles of active video grounding, external retrieval, and long-horizon reinforcement learning.

cs.CV↗

arXiv2026

Kwai Keye-VL-2.0 Technical Report

We introduce Kwai Keye-VL-2.0-30B-A3B, an open-source Mixture-of-Experts (MoE) multimodal foundation model designed to advance long-video understanding and agentic intelligence. To address the challenges of ultra-long contexts, information redundancy, and prohibitive computational costs inherent in hour-level videos, Keye-VL-2.0 is the first to adapt DeepSeek Sparse Attention (DSA) to GQA-based multimodal architectures, enabling lossless 256K context processing while capturing critical frames and long-range temporal dependencies. This architecture is underpinned by a highly optimized training and inference infrastructure, including scalable video I/O, heterogeneous ViT-LM parallelism, and custom DSA kernels that significantly maximize throughput and minimize computational overhead. Furthermore, to overcome the algorithmic dilemma of catastrophic forgetting during multi-task alignment, we introduce Cross-Modal Multi-Teacher On-Policy Distillation (MOPD) paired with Context-RL and Video-RL. By distilling dense token-level teacher feedback from on-policy rollouts back into the MoE backbone, which activates only 3B parameters, Keye-VL-2.0 natively empowers advanced agent collaboration across Code, Tool, and Search scenarios with multimodal self-correction. Extensive evaluations across video understanding, temporal grounding, reasoning, STEM, and agent benchmarks demonstrate that Keye-VL-2.0-30B-A3B achieves state-of-the-art performance among models of similar scale, particularly excelling in fine-grained temporal localization on TimeLens and long-video comprehension on Video-MME-v2 and LongVideoBench. We release our model checkpoints to accelerate community progress toward scalable and robust multimodal agentic applications.

cs.CV↗

arXiv2026

BiSegMamba: Efficient Bidirectional Tri-Oriented Mamba for 3D Medical Image Segmentation

Accurate 3D medical image segmentation requires both fine spatial detail and long-range volumetric context. Although Mamba provides efficient long-sequence modeling, volumetric formulations remain sensitive to spatial serialization order, repeated multi-orientation processing, and fixed aggregation of orientation-specific representations. We propose BiSegMamba, an efficient bidirectional tri-oriented Mamba framework that follows a compact-to-detail and local-to-global processing strategy. A progressive compacting stem shifts semantic reasoning to compact representations while retaining shallow high-resolution features for reconstruction, and a multi-scale spatial mixer captures local anatomical patterns. For global context modeling, volumetric features are serialized along three anatomical orientations and their reversed directions, the resulting six sequences are processed jointly by a shared Mamba operator, and reconciled using learned channel-wise forward-backward weighting. The restored orientation-specific features are subsequently integrated through input-dependent directional fusion. Experiments on an in-house carotid dataset and ACDC, BraTS2023, and AMOS-CT achieve Dice scores of 96.3%, 91.63%, 92.57%, and 89.03%, respectively. Compared with SegMamba-V2 under standardized $128^3$ profiling, BiSegMamba reduces the parameter count by 65.9% (138.77M$\rightarrow$47.38M) and inference latency by 53.6% (129.08$\rightarrow$59.86 ms), while increasing throughput by 115.7% (7.75$\rightarrow$16.71 samples/s). These results demonstrate a favorable accuracy--efficiency trade-off among volumetric Mamba-based segmentation models.

cs.CV↗

arXiv2026

Video-CoE: Reinforcing Video Event Prediction via Chain of Events

Despite advances in the application of MLLMs for various video tasks, video event prediction (VEP) remains relatively underexplored. VEP requires the model to perform fine-grained temporal modeling of videos and establish logical relationships between videos and future events, which current MLLMs still struggle with. In this work, we first present a comprehensive evaluation of current leading MLLMs on the VEP task, revealing the reasons behind their inaccurate predictions, including lack of logical reasoning ability for future events prediction and insufficient utilization of visual information. To address these challenges, we propose \textbf{C}hain \textbf{o}f \textbf{E}vents (\textbf{CoE}) paradigm, which constructs temporal event chains to implicitly enforce MLLM focusing on the visual content and the logical connections between videos and future events, incentivizing model's reasoning capability with multiple training protocols. Experimental results on public benchmarks demonstrate that our method outperforms both leading open-source and commercial MLLMs, establishing a new state-of-the-art on the VEP task. Codes and models will be released soon.

cs.CV↗

arXiv2026

VideoTemp-o3: Harmonizing Temporal Grounding and Video Understanding in Agentic Thinking-with-Videos

In long-video understanding, conventional uniform frame sampling often fails to capture key visual evidence, leading to degraded performance and increased hallucinations. To address this, recent agentic thinking-with-videos paradigms have emerged, adopting a localize-clip-answer pipeline in which the model actively identifies relevant video segments, performs dense sampling within those clips, and then produces answers. However, existing methods remain inefficient, suffer from weak localization, and adhere to rigid workflows. To solve these issues, we propose VideoTemp-o3, a unified agentic thinking-with-videos framework that jointly models video grounding and question answering. VideoTemp-o3 exhibits strong localization capability, supports on-demand clipping, and can refine inaccurate localizations. Specifically, in the supervised fine-tuning stage, we design a unified masking mechanism that encourages exploration while preventing noise. For reinforcement learning, we introduce dedicated rewards to mitigate reward hacking. Besides, from the data perspective, we develop an effective pipeline to construct high-quality long video grounded QA data, along with a corresponding benchmark for systematic evaluation across various video durations. Experimental results demonstrate that our method achieves remarkable performance on both long video understanding and grounding.

cs.CV↗

arXiv2025

EventFormer: A Node-graph Hierarchical Attention Transformer for Action-centric Video Event Prediction

Script event induction, which aims to predict the subsequent event based on the context, is a challenging task in NLP, achieving remarkable success in practical applications. However, human events are mostly recorded and presented in the form of videos rather than scripts, yet there is a lack of related research in the realm of vision. To address this problem, we introduce AVEP (Action-centric Video Event Prediction), a task that distinguishes itself from existing video prediction tasks through its incorporation of more complex logic and richer semantic information. We present a large structured dataset, which consists of about $35K$ annotated videos and more than $178K$ video clips of event, built upon existing video event datasets to support this task. The dataset offers more fine-grained annotations, where the atomic unit is represented as a multimodal event argument node, providing better structured representations of video events. Due to the complexity of event structures, traditional visual models that take patches or frames as input are not well-suited for AVEP. We propose EventFormer, a node-graph hierarchical attention based video event prediction model, which can capture both the relationships between events and their arguments and the coreferencial relationships between arguments. We conducted experiments using several SOTA video prediction models as well as LVLMs on AVEP, demonstrating both the complexity of the task and the value of the dataset. Our approach outperforms all these video prediction models. We will release the dataset and code for replicating the experiments and annotations.

cs.CV↗

arXiv2025

Fold-transversal surface code cultivation

Magic state cultivation is a state-of-the-art protocol to prepare ultra-high fidelity non-Clifford resource states for universal quantum computation. It offers a significant reduction in spacetime overhead compared to traditional magic state distillation techniques. Cultivation protocols involve measuring a transversal logical Clifford operator on an initial small-distance code and then rapidly growing to a larger-distance code. In this work, we present a new cultivation scheme in which we measure the fold-transversal Hadamard of the unrotated surface code, and leverage unitary techniques to grow within the surface code family. Using both stabilizer and state vector simulations we find that this approach achieves the lowest known spacetime overhead for magic state cultivation. Practical implementation of our protocol is best suited to architectures with non-local connectivity, showing the strength of architectures where such connectivity is readily available.

quant-ph↗

arXiv2025

Taming coherent noise with teleportation

Compared to the more widely studied Pauli errors, coherent errors present several new challenges in quantum computing and quantum error correction (QEC). For example, coherent errors may interfere constructively over a long circuit and significantly increase the overall failure rate compared to Pauli noise. Additionally, there is so far no analytical proof for a topological code threshold under coherent errors. Moreover, it is hard to even numerically estimate the performance of QEC under coherent errors as their effect in a Clifford circuit cannot be efficiently classically simulated. In this work, we demonstrate that teleportation effectively tailors coherent errors into Pauli errors, for which analytical and numerical results are abundant. We first show that repeated teleportation of a single qubit decoheres errors, and the average infidelity grows at worst linearly with the number of teleportations, similar to Pauli errors. We then analyze a physically motivated pure $Z$-coherent error model for teleported CSS codes in which over-rotation errors accompany every gate, and find that such an error model is equivalent to a Pauli error model. Our result implies that the performance of a CSS code implemented via teleportation-based error correction or measurement-based error correction with such coherent noise can be efficiently simulated on a classical computer and has an analytically provable threshold. The intrinsic noise-tailoring property of teleportation may ultimately remove the need for randomized compiling in teleportation-based quantum computing schemes.

quant-ph↗

arXiv2025

VidEvent: A Large Dataset for Understanding Dynamic Evolution of Events in Videos

Despite the significant impact of visual events on human cognition, understanding events in videos remains a challenging task for AI due to their complex structures, semantic hierarchies, and dynamic evolution. To address this, we propose the task of video event understanding that extracts event scripts and makes predictions with these scripts from videos. To support this task, we introduce VidEvent, a large-scale dataset containing over 23,000 well-labeled events, featuring detailed event structures, broad hierarchies, and logical relations extracted from movie recap videos. The dataset was created through a meticulous annotation process, ensuring high-quality and reliable event data. We also provide comprehensive baseline models offering detailed descriptions of their architecture and performance metrics. These models serve as benchmarks for future research, facilitating comparisons and improvements. Our analysis of VidEvent and the baseline models highlights the dataset's potential to advance video event understanding and encourages the exploration of innovative algorithms and models. The dataset and related resources are publicly available at www.videvent.top.

cs.CV↗

arXiv2024

Unraveling the switching dynamics in a quantum double-well potential

The spontaneous switching of a quantum particle between the wells of a double-well potential is a phenomenon of general interest to physics and chemistry. It was broadly believed that the switching rate decreases steadily with the size of the energy barrier. This view was challenged by a recent experiment on a driven superconducting Kerr nonlinear oscillator (often called the Kerr-cat qubit or the Kerr parametric oscillator), whose energy barrier can be increased by ramping up the drive. Remarkably, as the drive amplitude increases, the switching rate exhibits a step-like decrease termed the "staircase". The view challenged by the experiment demands a deep review of our understanding of quantum effects in double wells. In this work, we derive a semi-analytical formula for the switching rate that resolves a continuous transition between tunneling- and dissipation-dominated dynamics. These two dynamics are observed respectively in the flat and the steep parts of each step in the staircase. Our formula exposes two distinct dissipative processes that limit tunneling: dephasing and decay. This allows us to predict the critical drive amplitudes where steps occur. In addition, we show that in the regime of a few states in the well and under moderate to low temperatures, highly excited states are populated predominantly via cascaded and direct thermal heating rather than quantum heating. At very low temperatures, however, the perturbation induced by the nonhermitian Hamiltonian becomes important and facilitates a new form of quantum heating. We numerically map the activation mechanism as a function of drive amplitude, damping rate, and temperature. Our theory deepens the understanding of switching dynamics between metastable quantum states, highlights the importance of a general interplay between tunneling and dissipation, and identifies a novel quantum regime in activated transitions.

quant-ph↗

arXiv2022

The squeezed Kerr oscillator: spectral kissing and phase-flip robustness

By applying a microwave drive to a specially designed Josephson circuit, we have realized an elementary quantum optics model, the squeezed Kerr oscillator. This model displays, as the squeezing amplitude is increased, a cross-over from a single ground state regime to a doubly-degenerate ground state regime. In the latter case, the ground state manifold is spanned by Schr\"odinger-cat states, i.e. quantum superpositions of coherent states with opposite phases. For the first time, having resolved up to the tenth excited state in a spectroscopic experiment, we confirm that the proposed emergent static effective Hamiltonian correctly describes the system, despite its driven character. We also find that the lifetime of the coherent state components of the cat states increases in steps as a function of the squeezing amplitude. We interpret the staircase pattern as resulting from pairwise level kissing in the excited state spectrum. Considering the Kerr-cat qubit encoded in this ground state manifold, we achieve for the first time quantum nondemolition readout fidelities greater than 99%, and enhancement of the phase-flip lifetime by more than two orders of magnitude, while retaining universal quantum control. Our experiment illustrates the crucial role of parametric drive Hamiltonian engineering for hardware-efficient quantum computation.

quant-ph↗