Proposes streaming detection of AI-generated videos from compressed bitstream using motion field analysis, achieving anytime-valid inference with five orders of magnitude lower compute than pixel CNNs.

Mert Onur Cakiroglu, Mehmet Dalkilic, Hasan Kurban

This paper introduces Riemannian flow matching in VGGT's latent space for 3D scene generation, outperforming baselines with improved geometric consistency.

Lisa Weijler, Irene Ballester, Guofeng Mei et al.

COPD enhances student model reasoning efficiency by using contrastive teaching signals to reduce reasoning length.

Jiacheng Ruan, Jun Tang, Wenzhen Yuan et al.

Self-SiMS uses video self-similarity for zero-shot proposals, reaching 42.2 [email protected] on QVHighlights test.

Jihyun Lee, Cheol-Ho Cho, Woojin Jun et al.

This paper introduces a perception-guided framework for single-view 3D reconstruction, achieving up to 32.7% reduction in Chamfer distance and significant metric improvements.

Y Huynh, Duc Thanh Nguyen, Mohamed Abdelrazek

Proposes Plenoptic Condensation (PCon), a multi-stage scene reconstruction method achieving sub-millimeter accuracy and detailed reflectance modeling.

Brevin Tilmon, Alex DeJournett, John Leffingwell et al.

SciForma combines M-DPO with structural inventories to improve component, arrow, and text fidelity, outperforming open-source baselines and GPT-Image-1.5.

Yuxuan Luo, Peng Zhang, Xinjie Zhang et al.

Future-state-conditioned VLN (FSC-VLN) introduces training-only future supervision via dual queries, boosting long-horizon navigation success by 3-4% on R2R.

Lingfeng Zhang, Zhanguang Zhang, Liheng Ma et al.

MoGe-3 improves monocular geometry estimation accuracy using Self-Guided Sparse Volumetric Refinement (SSR), solving thin structure distortion issues.

Lingyu Kong, Ruicheng Li, Ruicheng Wang et al.

FF-ProCams uses feed-forward Gaussian modeling with Mamba2-Transformer to achieve real-time multi-view scene reconstruction in projector-camera systems.

Ziyao Wang, Yuqi Li, Wenxing Zheng et al.

ShotPlan introduces learnable planning tokens with FRoPE for frame-level shot control, outperforming existing cinematic video methods.

Su Guo, Guangce Liu, Haosen Yang et al.

ConsiSpace leverages geometric consistency to enhance video spatial reasoning, improving stability and efficiency with a 12.6-point gain.

Ting Huang, Zhenyu Zhang, Wenyuan Huang et al.

TimeLens2 leverages multimodal LLMs with Wasserstein-based rewards for multi-interval video grounding, outperforming models up to 397B parameters.

Yuhan Zhu, Changlian Ma, Xiangyu Zeng et al.

EditCLEVR: A benchmark for evaluating object-centric representations under semantic edits.

Anuraag Gadehothur Karnam, Tarunesh Sathish

ParaCAD employs a surface-centric autoregressive model to directly generate native parametric B-Rep surfaces conditioned on point clouds, improving geometric accuracy.

Dafei Qin, Rui Xu, Zeyu Shen et al.

Proposed CCS framework disrupts both understanding and generation branches in UMMs, reducing facial identity preservation from high to zero, with superior robustness.

Weiwei Tan, Junxian Li, Rui Wang et al.

Scene-SAM3D improves multi-view 3D scene generation without fine-tuning, reducing CD by 43.8% on Replica.

Yuqi Zhang, Yadan Luo, Xiangyu Sun et al.

Proposes PAVXploreRL, a reward-driven RL framework optimizing physical plausibility, action adherence, and visual fidelity, improving world model robustness by 5.6%.

Han Wang, Zijun Wang, Shuoshuo Xue et al.

Category-agnostic 3D editing framework using Coupled Neural Shape (CNS) with joint optimization of latent codes and neural feature volumes.

Jingyu Hu, Weilong Yan, Zhengzhe Liu et al.

Proposes IoU-PD, leveraging true bounding boxes as privileged training guidance, significantly improving multimodal large model's visual grounding accuracy by ~4.25% in mIoU.

Xiuyuan Zhu, Ke Lu, Hao Wu et al.