PhyS framework improves streaming world models' physical consistency by distilling physical priors, enhancing benchmarks like PhysicsIQ.

Liangliang Zhao, Junying Wang, Danni Yang et al.

FlexSplat introduces a calibration-free 3D Gaussian splatting framework using joint depth and camera prediction, achieving near state-of-the-art results without known poses.

Amir Sabbaghziarani, Hanting Ye, Maria Gorlatova et al.

Proposes ICQ task, builds ISYV dataset and model, achieving 57% accuracy on person-centric video reasoning, improving long-horizon and cross-domain identification.

Shibo Gao, Chongxiao Wang, Chenglong Huang et al.

UniJEPA unifies image and video predictive modeling in a single latent space, using a single end-to-end loss with Gaussian regularization, enabling task-agnostic world modeling and zero-shot planning.

An Lanji, Dawei Liu, Jin Li et al.

GeoDistill-Refine employs multi-prompt fusion and signed distance fields, boosting annotation-free spacecraft segmentation IoU by 4.56% and Boundary F1 by 13.80%.

Yonglong Zhang, Zongwu Xie, Yang Liu

CANIS uses a generation-assisted semantic bridge for category-agnostic 3D canonicalization, improving classification and segmentation performance.

Kendong Liu, Yuxin Yao, Junhui Hou

DCMPC-Net uses cross-modal prompts for unified adverse-weather restoration, but the supplied text reports no numerical metrics.

Wanshu Fan, Yunzhe Zhang, Yue Shen et al.

HSMLA combines multi-scale linear attention with content-aware sparse softmax, achieving 4× speedup with high accuracy in high-res vision tasks.

Dong Liu, Yanxuan Yu, Renata Borovica-Gajic et al.

Gated Hindsight Distillation (GHD) leverages next screenshots as privileged training info, significantly improving mobile GUI agent success rates.

Weiwei Li, Junzhuo Liu, Tong Chu et al.

Vorch-Director matches residual corrections to flow-matching noise levels, improving long-horizon audio-visual stability; no numeric scores are provided.

Lisai Zhang, Yidi Wu, Qi Liu et al.

Proposes an iterative hybrid discrete-continuous viewpoint planning method to enhance UAV photogrammetry reconstruction accuracy and completeness.

Alan Grech, Daniel Pisani, Andre Grima et al.

Vorch-Streamer combines mixed Teacher Forcing and Diffusion Forcing with long-horizon Self Forcing, achieving 27.12 FPS real-time long-form audio-video generation.

Menglin Han, Yang Ding, Yulei Lu et al.

Vorch-IR is a unified multimodal video editing framework supporting multi-person identity and background replacement, built on LTX2 with attention mechanisms.

Yaole Wang, Xiaoyu Chen, Xin Ma et al.

Proposes CoCo-IR and TIE model, leveraging large multimodal models for multi-turn contextual image retrieval; achieves 39.4 mAP@5 (single-turn) and 44.1 R@1 (4-turn).

Shengcao Cao, Tanmaya Shekhar Dabral, Zhongli Ding et al.

AV-MSF combines multi-view images and few impact recordings to physically reconstruct object impact sounds, outperforming physics-based and data-driven baselines.

Zisen Shao, Zihao Wei, Derong Jin et al.

Leveraging foundation vision models for cross-species animal pose tracking, achieving high accuracy with limited labels.

Le Li, Daniela Ivanova, Nicolas Pugeault

MAGIC combines graph label propagation and geometric alignment to stabilize feature space in semi-supervised class-incremental learning, reducing drift and improving accuracy.

Yousef Abdi, Mohammad Asadpour, Yousef Seyfari

ToolArtist employs post-trained UMM with RL and RAD-GRPO to dynamically coordinate reasoning, tool use, and image generation, outperforming fixed pipeline methods.

Jiahao Zhao, Xiaomin Yu, Zhongxiang Sun et al.

Introduces CLIP-CC-Bench, a multi-model ensemble framework for evaluating paragraph-level video descriptions, covering 17 SOTA models with 200 movie clips.

Mukhtiar Ali, Harsh Dubey, Sugam Mishra et al.

DyLaR enhances video QA accuracy to 58.2% with under 20 tokens per query using dynamic latent reasoning.

Haotian Xia, Zilin Xiao, Junbo Zou et al.