VideoAgent2 employs uncertainty-aware chain-of-thought reasoning with plan adjustment, achieving 13.1% average improvement on long video benchmarks.

Zhuo Zhi, Qiangqiang Wu, Minghe shen et al.

JarvisIR employs VLM-controlled multi-expert restoration models, achieving 50% improvement in perception metrics under adverse weather, enhancing autonomous driving robustness.

Yunlong Lin, Zixu Lin, Haoyu Chen et al.

ScreenSeekeR uses hierarchical reasoning and visual search to boost GUI grounding accuracy to 48.1% in high-res professional settings.

Kaixin Li, Ziyang Meng, Hongzhan Lin et al.

Introduced RISEBench, a benchmark for reasoning-based visual editing; top model GPT-4o-Image scores only 28.8%, highlighting significant gaps.

Xiangyu Zhao, Peiyuan Zhang, Kexian Tang et al.

SkyReels-A2 employs video diffusion transformers with multi-modal joint embedding to generate controllable videos with high element fidelity.

Zhengcong Fei, Debang Li, Di Qiu et al.

Proposes WoTE framework using BEV world model for online trajectory evaluation, achieving state-of-the-art results.

Yingyan Li, Yuqi Wang, Yang Liu et al.

Ross3D integrates cross-view and global reconstruction tasks into visual instruction tuning, achieving SOTA with 63.0 EM on SQA3D and 107.0 CIDEr on ScanQA.

Haochen Wang, Yucheng Zhao, Tiancai Wang et al.

SpaceR leverages RLVR and map imagination, with 151k data, to enhance video spatial reasoning, outperforming GPT-4o by 11.6%.

Kun Ouyang, Yuanxin Liu, Haoning Wu et al.

SuperDec employs superquadric primitives for compact 3D scene decomposition, trained on ShapeNet, generalizes well to ScanNet++ and Replica datasets.

Elisabetta Fedele, Boyang Sun, Leonidas Guibas et al.

KOFFVQA: An objectively evaluated free-form VQA benchmark for Korean language VLMs.

Yoonshik Kim, Jaeyoon Jung

DiT4SR integrates bidirectional attention and local info injection, significantly improving real-world image super-resolution performance.

Zheng-Peng Duan, Jiawei Zhang, Xin Jin et al.

Proposed Magiv3 unifies detection, OCR, and grounding tasks, enabling automated manga-to-text conversion with 85% detection accuracy and 92% OCR rate.

Ragav Sachdeva, Andrew Zisserman

Proposes Enhanced Instance Replay (EIR) combining dynamic class merging and background fusion to mitigate catastrophic forgetting in continual semantic segmentation.

Hongmei Yin, Tingliang Feng, Fan Lyu et al.

Introduces Video-R1 with T-GRPO algorithm and hybrid datasets, achieving 37.1% accuracy on VSI-Bench, surpassing GPT-4o.

Kaituo Feng, Kaixiong Gong, Bohao Li et al.

LeX-Art enhances text rendering in T2I models via high-quality data synthesis, achieving 79.81% PNED improvement and superior attribute accuracy.

Shitian Zhao, Qilong Wu, Xinyue Li et al.

Wan employs a diffusion transformer with 14B parameters, achieving state-of-the-art video generation performance.

Team Wan, Ang Wang, Baole Ai et al.

Introduces VILBench and 73.6K vision-language process reward data, enabling training of a 3B PRM that outperforms baselines by 3.3%.

Haoqin Tu, Weitao Feng, Hardy Chen et al.

DINeMo leverages large pretrained models for pseudo-correspondence, enabling zero-shot 3D pose estimation with 67.3% gap reduction to supervised methods.

Weijie Guo, Guofeng Zhang, Wufei Ma et al.

AVUT benchmark evaluates audio-centric video understanding, introducing multi-task tasks and a permutation filtering mechanism to prevent text shortcuts.

Yudong Yang, Jimin Zhuang, Guangzhi Sun et al.

OpenFunGraph leverages foundation models to generate functional 3D scene graphs, outperforming Open3DSG and ConceptGraph.

Chenyangguang Zhang, Alexandros Delitzas, Fangjinhua Wang et al.