The paper proposes a training-free framework, Visual Representation-Guided Video-LLM Reasoning, to perform composed video retrieval by using visual examples and text instructions, achieving strong per…

The paper proposes a zero-shot reason-then-retrieve pipeline using Qwen3.5-27B to solve the challenging task of composed video retrieval (CoVR-R), achieving high performance on both validation and bli…

The paper introduces ConTrans, a novel local-global multi-scale encoder that combines convolutional and transformer features to significantly improve zero-shot temporal action localization by capturin…

The paper introduces VIP-Net, a framework that leverages multi-modal spatio-temporal cues and a new dataset (Temporal-VIP) to accurately identify the most influential people in videos, overcoming the…

The paper introduces a novel two-stage framework to achieve robust, category-agnostic object localization in-context (ICL) by optimizing attention and minimizing localization error using reinforcement…

The paper proposes ST-DRC, a Spatial-Temporal Decoupled Reference Conditioning framework that effectively balances high-level semantic control and low-level identity fidelity for text-to-video generat…

LongLive-RAG proposes a novel Retrieval-Augmented Generation (RAG) framework to stabilize and improve the quality of long-horizon video generation by treating the entire generated history as a searcha…

The paper proposes Visual Gradient Steering (VGS), a method that decomposes the distillation loss into language and visual components and steers the optimization to prioritize visual grounding, signif…

The paper introduces Moment-Video, a new benchmark that diagnoses the ability of video MLLMs to understand brief, critical visual events, revealing that current models struggle significantly with temp…

CIPER proposes a unified transformer framework to simultaneously perform cross-view image retrieval and precise 3-DoF pose estimation, overcoming the limitations of cascaded, separate methods.

The paper introduces Q-ALIGN DT, a novel framework that improves conditioned sequence models by enforcing alignment between the input return-to-go (RTG) signal and the output policy's expected Q-value…

TunerDiT introduces a training-free progressive steering method to enhance multi-event video generation using Diffusion Transformers, achieving state-of-the-art performance by explicitly managing even…

The paper introduces the Terminal Representation (TR), a novel, lower-dimensional, and structurally distinct formulation for encoding reward-weighted trajectories in RL that bypasses the need for eige…

The paper introduces BabyCL, a continual multimodal learning framework that processes egocentric video data in a single chronological pass, demonstrating that meaningful word-referent mappings can be…

The paper proposes EKSFT, a selective fine-tuning method that masks high-entropy or high-KL divergence tokens during Supervised Fine-Tuning (SFT) to prevent distribution shift and improve subsequent R…

VideoMLA introduces a novel Multi-Head Latent Attention (MLA) mechanism that replaces per-head KV caches with a shared low-rank content latent, significantly reducing memory and improving throughput f…

肖代替了视觉令牌的永久删除，通过可恢复的路由来改进视觉语言模型的性能