ARC-CT employs anatomy-guided contrastive learning, combining regional features to achieve 0.86 macro AUC on chest CT abnormalities.

Huseyin Umut Isik, Mehmet Alp Ozaydin, Sila Kurugol et al.

Scaling laws for video diffusion models in driving data show validation loss follows power laws; a 9B-parameter model predicts loss of ~0.0753 with 3.6% error.

Victor Besnier, Anh-Quan Cao, Elias Ramzi et al.

StreamEMS enhances video streaming understanding in vision-language models with a self-evolving memory scheme, excelling on OVO-Bench.

Yuxin Liu, Peiqin Zhuang, Yali Wang

VGTok optimizes high-resolution object detection using regional separability, achieving 48.38 AP on VisDrone dataset.

Khayrul Islam

This paper introduces MILO, leveraging Large Reconstruction Models (LRMs) to reconstruct detailed 3D human-object interactions from a single image, outperforming state-of-the-art methods.

Agniv Chatterjee, Georgios Pavlakos

LeVJEPA uses collapse-free invariance loss and SIGReg regularization, reducing pretraining compute by up to 20x while maintaining or surpassing state-of-the-art performance.

Lukas Kuhn, Lucas Maes, Giuseppe Serra et al.

Proposed TADP: task-aware deformable prediction for single-stage 3D detection, achieving 80.91% mAP on KITTI.

Su Wang, Yaochen Li, Min Yang et al.

DINOcular integrates depth priors with self-supervised vision transformers, boosting 3D spatial understanding by 15% on Probe3D benchmarks.

Farkhat Almukhamedov, Sami Azirar, Hermann Blum

PACE achieves 93.8% performance retention while reducing visual tokens by 90%, enabling a 3.1x speedup via pixel compression and dual-attention extraction.

Junjie Liu, Shengyuan Ye, Xu Chen

Video-FLAIR uses reinforcement learning to select reasoning modes, improving accuracy and reducing computation.

Yogesh Kulkarni, Pooyan Fazli

VBVR-Pro introduces 300 procedurally generated tasks, rule-based reward scorers, and multi-modal evaluation, advancing scalable and reliable visual reasoning.

Junxiang Xu, Ruisi Wang, Fanyi Pu et al.

VDA framework with VC-OT and VMA effectively mitigates visual dependence drift in unsupervised continual multimodal learning, achieving 62.5% accuracy across six tasks.

Kaichen Li, Zhilin Zhu, Jianhao Huang et al.

StreamPI employs instruction-anchored streaming multimodal temporal modeling, leveraging pre-trained LLM length extrapolation for parameter-free multi-frame inference, outperforming pi0.5 in robot tasks.

Zhe Liu, Jinghua Hou, Yuxiang Lu et al.

4DStreamCtrl unifies 3D point-track representation for real-time video generation and control, enhancing motion precision.

Shiqian Li, Chenguo Lin, Zhiguang Liu et al.

MTAR framework achieves efficient autoregressive image generation on ImageNet with multi-token prediction, reducing FID by 0.95.

Guo Niu, Xiongfei Yao, Teng Wang et al.

ProViP enhances VLM inference efficiency via head-aware pruning, retaining 95.9% performance with 1.62x speedup.

Chaofang Ma, Lin Jiang, Carol Jingyi Li et al.

V-Link restores visual cues in Action DiT, improving GR00T N1.6 by 31.2% on LIBERO-Plus and 18.8% on RoboTwin 2.0.

Yehao Lu, Jiarui Yang, Yuning Su et al.

Unified framework models smart glasses as closed-loop systems with eight hardware capability axes, connecting perception, state, and action.

Jiangning Zhang, Haojun Chen, Yong Liu

LAION-BVD is a 10M-hour open video dataset with 1.3B URLs, supporting large-scale multimodal pretraining with content-aware scene detection.

Andreas Hochlehnert, Marianna Nezhurina, Mehdi Cherti et al.

Proposed MoE-based feature adapter for prompt-free binary coronary artery segmentation, significantly improving cross-dataset generalization.

Lin Xi, Yingliang Ma