Deep metric learning with depth data achieves 99% accuracy in cattle identification, eliminating the need for breed-specific markings.

Asheesh Sharma, Lucy Randewich, William Andrew et al.

ECLIPSE leverages Visual Prompt Tuning, freezing backbone parameters and fine-tuning prompts, achieving state-of-the-art continual panoptic segmentation with only 1.3% trainable parameters.

Beomyoung Kim, Joonsang Yu, Sung Ju Hwang

MagicLens uses self-supervised learning on 36.7M web triplets to support open-ended image retrieval, outperforming SOTA.

Kai Zhang, Yi Luan, Hexiang Hu et al.

Introduced inconsistency-guided detail regularization, enhancing mask-guided matting accuracy, surpassing SOTA methods.

Weihao Jiang, Zhaozhi Xie, Yuxiang Lu et al.

ObjectDrop uses counterfactual datasets and diffusion model fine-tuning to achieve photorealistic object removal and insertion, outperforming prior methods.

Daniel Winter, Matan Cohen, Shlomi Fruchter et al.

Mini-Gemini employs dual visual encoders and patch info mining to enhance high-res visual understanding, surpassing private models in zero-shot benchmarks.

Yanwei Li, Yuechen Zhang, Chengyao Wang et al.

Proposed a two-stage diffusion-based framework utilizing scene affordance for language-guided human motion synthesis, outperforming baselines on benchmark datasets.

Zan Wang, Yixin Chen, Baoxiong Jia et al.

Proposed GoodSAM framework combines DAR and MKA modules to transfer SAM's zero-shot instance segmentation for panoramic semantic segmentation, achieving +3.75% mIoU improvement.

Weiming Zhang, Yexin Liu, Xu Zheng et al.

PruMerge adaptively prunes and merges visual tokens, achieving 14× compression while maintaining performance.

Yuzhang Shang, Mu Cai, Bingxin Xu et al.

Introduced MathVerse, a multi-version benchmark with chain-of-thought analysis, revealing most models rely heavily on text rather than visual understanding in math problems.

Renrui Zhang, Dongzhi Jiang, Yichi Zhang et al.

Unsupervised cross-view geo-localization using projection and re-ranking achieves over 40% pseudo-label accuracy.

Guopeng Li, Ming Qian, Gui-Song Xia

mPLUG-DocOwl 1.5 improves document understanding via Unified Structure Learning, achieving >10-point SOTA gains on 5/10 benchmarks.

Anwen Hu, Haiyang Xu, Jiabo Ye et al.

SpatialPIN enhances VLM's 3D reasoning via multi-model prompting, improving spatial VQA and robotics tasks without training.

Chenyang Ma, Kai Lu, Ta-Ying Cheng et al.

VideoAgent integrates structured memory with foundation models, boosting long video understanding by 6.6% on NExT-QA.

Yue Fan, Xiaojian Ma, Rujie Wu et al.

RebQ framework improves average precision from 20.00 to 50.92 and reduces forgetting from 75.95 to 8.56 in missing modality learning.

Shu Zhao, Xiaohan Zou, Tan Yu et al.

DiffUIR employs a selective hourglass mapping strategy with strong condition guidance and shared distribution terms, achieving state-of-the-art performance in multi-task image restoration with only 0.89M parameters.

Dian Zheng, Xiao-Ming Wu, Shuzhou Yang et al.

VideoAgent uses a large language model as an agent for multi-round retrieval, achieving 54.1% zero-shot accuracy with only 8.4 frames on average.

Xiaohan Wang, Yuhui Zhang, Orr Zohar et al.

Proposes MM1, a multimodal pretraining framework combining contrastive and reconstructive losses, achieving SOTA with 30B parameters.

Brandon McKinzie, Zhe Gan, Jean-Philippe Fauconnier et al.

OpenGraph introduces a hierarchical 3D graph framework using vision-language models for large-scale outdoor scene understanding, achieving 73.4% IoU without fine-tuning.

Yinan Deng, Jiahui Wang, Jingyu Zhao et al.

Proposes TRUMANS dataset and diffusion-based autoregressive model for arbitrary-length human-scene interaction motion synthesis.

Nan Jiang, Zhiyuan Zhang, Hongjie Li et al.