Explore · 每周精选
发现最受关注的研究论文,追踪研究趋势,订阅感兴趣的期刊与关键词。
42 篇论文 · 按点赞排序
Dahun Kim, Anelia Angelova, Weicheng Kuo
CFM-ViT enhances open-vocabulary object detection and image-text retrieval through combined masked and contrastive learning objectives, positional embedding dropout, and region-level representation learning.
Jensen Gao, Bidipta Sarkar, Fei Xia +5 authors
A physically-grounded vision-language model (VLM) fine-tuned on a dataset of object physical properties improves robotic manipulation and planning performance.
Erik Nijkamp, Tian Xie, Hiroaki Hayashi +22 authors
XGen, a series of open-source 7B parameter models trained on up to 8K sequence lengths, outperforms other open-source LLMs in long sequence modeling tasks.
Marko Mihajlovic, Sergey Prokudin, Marc Pollefeys +1 authors
ResFields, a neural network variant with temporal residual layers, enhance the representation of complex temporal signals in neural fields, improving performance across 2D video approximation, dynamic shape modeling, and NeRF reconstruction.
David B. D'Ambrosio, Jonathan Abelian, Saminda Abeyruwan +32 authors
A robotic learning system for table tennis integrates perception, control, simulation, and real-world reset mechanisms to achieve precise ball returns and autonomous training with minimal latency.
Lily Goli, Cody Reading, Silvia Selllán +2 authors
BayesRays is a post-hoc framework that evaluates uncertainty in pre-trained NeRFs using spatial perturbations and Bayesian Laplace approximation, improving key metrics and applications.
Jiaxi Gu, Shicong Wang, Haoyu Zhao +7 authors
A Reuse and Diffuse framework (VidRD) enhances text-to-video generation by iteratively adding frames to initial video clips using latent features and temporal consistency enhancements.
Zhutian Yang, Jiayuan Mao, Yilun Du +4 authors
A compositional diffusion model learns to solve continuous constraint satisfaction problems by combining energies of diffusion models for individual constraints, enabling generalization and integration into task and motion planning.
Hanshu Yan, Jun Hao Liew, Long Mai +2 authors
MagicProp combines image editing and motion-aware appearance propagation using a diffusion-based model to modify video appearance while preserving motion consistency.
Marin Vlastelica, Tatiana López-Guevara, Kelsey Allen +3 authors
Denoising diffusion models combined with a particle sampling algorithm efficiently solve inverse design problems in fluid dynamics with reduced simulator calls.
Yuanpei Chen, Chen Wang, Li Fei-Fei +1 authors
A reinforcement learning system, Sequential Dexterity, chains dexterous policies to achieve complex, long-horizon manipulation tasks, demonstrating generalization and zero-shot transfer to real-world robots.
Vittorio Caggiano, Sudeep Dasari, Vikash Kumar
Multi-task learning is used to develop a manipulative prior for human-like dexterity, enabling agents to quickly learn a wide range of complex manipulation tasks using a realistic human hand model.
北京市昌平区探索星信息技术及软件开发工作室
京ICP备2026059466号