CameraAnything integrates pixel-level Plücker rays and resolution-aware 3D RoPE in a diffusion framework for joint camera parameter control, supporting viewpoint, focal length, and resolution adjustments.

Yixuan Li, Yanhong Zeng, Ka Leong Cheng et al.

Selective tuning of the last three ViT layers with HYBRID16 gives a 2B model 68.8% temporal mIoU from 16 frames.

Jiameng Zhang, Srikanth Madikeri

MSVS-VAE employs hierarchical point-shuffle upsampling and local AVS-Conv to achieve high-fidelity 3D reconstruction, 10× faster decoding, and 10× more compactness.

Dehao Hao, Kaiyi Zhang, Tanghui Jia et al.

TaoMate employs anchor-guided persistent memory, enabling stable real-time long-form audio-video generation with 97.55% appearance consistency and 16.32FPS speed.

Qijun Gan, Chenwei Zhang, Meiguang Jin et al.

PlanCraft integrates progressive sketches and spatial constraints, reducing FID by 61.1% in residential scene generation.

Pengyu Zeng, Yuqin Dai, Jun Yin et al.

UltraViT optimizes vision encoders for on-device large models, achieving 1.7x speed improvement.

Ioannis Maniadis Metaxas, Adrian Bulat, Alberto Baldrati et al.

SCALE uses self-supervised layout text generation and rule-guided sampling to boost sub-2nm local DRV fixing accuracy up to 97%.

Chia-Tung Ho, Haoyu Yang, Guanglei Zhou et al.

Proposes Trans-Unet, combining 3D-to-2D mapping, CNN, and self-attention, achieving high-fidelity brain folding prediction from point clouds.

Geran Zhao, Xiaotian Li, Poorya Chavoshnejad et al.

Introduces ConVBench for complex visual reasoning; uses GRPO reinforcement learning with consistency rewards to improve LVLM logical stability, achieving 73.36% consistency.

Liqiang Jing, Xiong Zhou, Siddharth Varia et al.

DAPM model achieves UAV monocular depth estimation from any height, pitch, roll, and FOV, reaching state-of-the-art performance.

Tong Ling, Wenhui Diao, Yingchao Feng et al.

Geo3R is a training-free geometric reasoning framework that reduces spatial hallucinations in multimodal large models, improving accuracy by over 10%.

Mingyu Wang, Weilin Jin, Wenbo Li et al.

Introduces ViSTR-Bench, a benchmark for evaluating spatial-temporal reasoning in dynamic scenes, covering Motion Perception, Spatial Relations, Outcome Prediction, and Physical Dynamics.

Han Li, Si Liu, Zehao Huang et al.

SubSplat uses SPGR to render high-res views from 256×256 inputs, reaching 25.52 dB on RealEstate10K.

Jiun Lee, Jaekwang Kim, Sangmin Lee

Axolotl3D unifies multi-modal inputs for faithful 3D shape completion under occlusion and multi-view conditions.

Anita Hu, Maria Shugrina

Proposes Masked Topology Modeling (MTM) for self-supervised learning on CAD B-rep graphs, achieving state-of-the-art results in geometric and topological understanding.

Heinrich Jiang, Jennifer Jang

PerceptDrive uses adaptive expert routing to fuse perception priors, achieving 90.4 PDMS on NAVSIM v1 without test-time candidate search.

Yushan Liu, Tianxiong Lv, Bohua Wang et al.

ENTRAP-VL uses a dual-modality probe to study contextual entrainment in vision-language models with a 1,500-item dataset.

Karan Goyal, Afreen Hossain, Debojyoti Das et al.

Introduces CAMA-BC with multi-layer multimodal alignment, achieving 78.5% F1 in backchannel prediction, especially for complex cues like empathy.

Min-Jae Kim, Jun-Yeong Moon, Mujeen Sung et al.

Trace environment uses scene grammar and executable task programs to enhance multi-domain visual reasoning, improving transferability by 3.5-4% on external benchmarks.

Md Tanvirul Alam

Proposes a multi-view panoptic segmentation method leveraging implicit geometric correspondence in large view synthesis models, achieving 33.56dB PSNR and 0.5949 mIoU without 3D reconstruction.

Kwonyoung Ryu, In-Jae Lee, Jonghyun Jin et al.