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。
50 篇论文 · 按点赞排序
Zeyuan Yang, Xueyang Yu, Delin Chen +2 authors
Mirage enhances vision-language models by integrating latent visual tokens into text decoding to improve multimodal reasoning without generating explicit images.
Ziyue Li, Chenrui Fan, Tianyi Zhou
Grokking, or continued test performance improvement after training loss convergence, is observed during pretraining of a large language model, showcasing a memorization-to-generalization process.
Yuqi Wang, Xinghang Li, Wenxuan Wang +5 authors
UniVLA is a multimodal VLA model that autoregressively processes vision, language, and action as token sequences, incorporating world modeling for effective long-horizon policy learning and achieving state-of-the-art results across simulation and real-world benchmarks.
Long Xing, Qidong Huang, Xiaoyi Dong +10 authors
ScaleCap enhances image captioning by iteratively enriching and calibrating captions using heuristic question answering and contrastive sentence rating, addressing multimodal and linguistic biases to improve accuracy, balance, and informativeness.
Yi Chen, Yuying Ge, Rui Wang +4 authors
GRPO-CARE, a reinforcement learning framework optimizing for consistency and correctness, outperforms standard GRPO on a new video understanding benchmark, SEED-Bench-R1, improving both performance and logical coherence in multimodal large language models.
Yukun Huang, Yanning Zhou, Jianan Wang +2 authors
Multi-plane synchronization extends 2D foundation models to 3D panorama generation, introducing DreamCube to achieve diverse appearances and accurate geometry.
Boyu Li, Siyuan He, Hang Xu +9 authors
A simulator named DualTHOR for training dual-arm humanoid robots integrates real-world assets and physics to enhance the robustness and generalization of Vision Language Models.
Jinyang Li, Xiaolong Li, Ge Qu +17 authors
A new benchmark and training environment for debugging SQL issues using advanced open-source models significantly improves their performance compared to proprietary solutions.
Runjia Li, Philip Torr, Andrea Vedaldi +1 authors
A novel memory mechanism called Surfel-Indexed View Memory enhances video generation by efficiently remembering and retrieving relevant past views, improving long-term scene coherence and reducing computational cost.
Sophia Sirko-Galouchenko, Spyros Gidaris, Antonin Vobecky +2 authors
A novel unsupervised post-training method improves dense image representations using pseudo-tasks and a pretrained diffusion model for in-context scene understanding.
Team Hunyuan3D, Shuhui Yang, Mingxin Yang +50 authors
The tutorial provides a comprehensive guide on using Hunyuan3D 2.1 for generating high-resolution, textured 3D models, covering data preparation, model architecture, training, evaluation, and deployment.
Tim Lawson, Laurence Aitchison
A novel conditional computation architecture for Transformers dynamically skips middle layers based on input and a gating mechanism, but does not outperform dense baselines in reducing computational cost or improving validation performance.
Tobias Vontobel, Seyedmorteza Sadat, Farnood Salehi +1 authors
HiWave enhances ultra-high-resolution image synthesis using pretrained diffusion models through a two-stage pipeline involving DDIM inversion and wavelet-based detail enhancement, improving visual fidelity and reducing artifacts.
Tianxing Chen, Zanxin Chen, Baijun Chen +23 authors
RoboTwin 2.0 is a scalable simulation framework for bimanual robotic manipulation that uses expert data synthesis and structured domain randomization to generate diverse and realistic synthetic data, improving sim-to-real transfer and generalization.
Co Tran, Salman Paracha, Adil Hafeez +1 authors
A preference-aligned routing framework using a compact 1.5B model effectively matches queries to user-defined domains and action types, outperforming proprietary models in subjective evaluation criteria.
Jianyun Xu, Song Wang, Ziqian Ni +4 authors
SAM4D is a multi-modal and temporal foundation model for segmentation in autonomous driving using Unified Multi-modal Positional Encoding and Motion-aware Cross-modal Memory Attention, with a multi-modal automated data engine generating pseudo-labels.
Jingwei Ni, Yu Fan, Vilém Zouhar +6 authors
LLMs struggle to predict human annotation disagreements, contrary to their performance in predicting majority labels, and RLVR-style reasoning exacerbates this issue.
Dylan Ebert
3D Arena evaluates generative 3D models using human preferences, revealing insights into visual and textural features' impact on quality.
Pranav Pawar, Akshansh Dwivedi, Jenish Boricha +2 authors
A new TTS architecture improves accent accuracy and emotion recognition for Hindi and Indian English by integrating phoneme alignment, culture-sensitive emotion embeddings, and dynamic accent code switching.
Minsoo Kim, Kyuhong Shim, Jungwook Choi +1 authors
InfiniPot-V is a training-free, query-agnostic framework that compresses the key-value cache during video encoding to maintain a fixed memory cap for streaming video understanding, enhancing real-time performance and accuracy.
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