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
Jiachen Li, Roberto Henschel, Vidit Goel +3 authors
MSG-VIM, a neural network for video instance matting, improves temporal consistency in alpha matte predictions using mask and feature guidance, outperforming existing methods on the VIM50 benchmark.
Gabriela Ben Melech Stan, Diana Wofk, Estelle Aflalo +4 authors
LDM3D-VR, comprising LDM3D-pano and LDM3D-SR, generates high-resolution RGBD from textual prompts and low-resolution inputs, demonstrating state-of-the-art performance in virtual reality applications.
Zhenfang Chen, Rui Sun, Wenjun Liu +2 authors
A generative neuro-symbolic visual reasoning model uses large language models to create and reuse modular code snippets for efficient visual reasoning.
Philip Andrew Mansfield, Arash Afkanpour, Warren Richard Morningstar +1 authors
A new family of local transformations using Gaussian random fields for image augmentations in self-supervised learning improves classification accuracy but requires careful tuning of hyperparameters.
George Kour, Marcel Zalmanovici, Naama Zwerdling +5 authors
A dataset called AttaQ is introduced to identify and evaluate vulnerable semantic regions in large language models, using adversarial questions to detect harmful responses and improve model safety.
Jianan Yao, Ziqiao Zhou, Weiteng Chen +1 authors
A prototype combining Large Language Models and static analysis reduces human effort in writing formal verification proof code by generating logical structures but overlooks context retention.
Junyan Li, Delin Chen, Yining Hong +4 authors
CoVLM enhances compositional reasoning in vision-language models by introducing communication tokens for iterative interaction between vision and language systems.
Hayeon Lee, Rui Hou, Jongpil Kim +4 authors
A novel co-training and co-distillation framework enhances performance and inference speed of smaller models through mutual knowledge transfer with larger models, outperforming traditional knowledge distillation methods.
Yanqin Jiang, Li Zhang, Jin Gao +2 authors
Consistent4D uses a 3D-aware image diffusion model and Cascade DyNeRF to generate 4D dynamic objects from uncalibrated monocular videos, improving spatial and temporal consistency with an Interpolation-driven Consistency Loss.
Sahana Ramnath, Brihi Joshi, Skyler Hallinan +6 authors
MaRio, a multi-reward conditioned self-rationalization algorithm, enhances the quality of rationales generated by small language models, improving both performance and the semantic properties of the rationales.
Sungho Jeon, Ching-Feng Yeh, Hakan Inan +4 authors
A simpler self-attention-only approach in pre-trained audio models matches the efficiency of more complex speech transformer models, especially with low-bit weight quantization.
Xiao Wang, Isaac Lyngaas, Aristeidis Tsaris +7 authors
The Long Short-Sequence Transformer (LSS Transformer) efficiently trains transformers on long sequences by segmenting them across GPUs with partial self-attention and novel communication techniques, achieving significant speed and memory improvements.
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