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。
Keivan Alizadeh, Iman Mirzadeh, Dmitry Belenko +5 authors
Efficient inference for large language models on devices with limited DRAM by optimizing data transfer and access from flash memory.
50 篇论文 · 按点赞排序
Akio Kodaira, Chenfeng Xu, Toshiki Hazama +7 authors
StreamDiffusion, a real-time diffusion pipeline, enhances interactive image generation by implementing batching denoising, residual classifier-free guidance, and stochastic similarity filtering, which boost throughput and reduce energy consumption.
Chi Zhang, Zhao Yang, Jiaxuan Liu +5 authors
A novel LLM-based multimodal agent learns to operate smartphone apps through autonomous exploration or imitation, demonstrating proficiency across diverse tasks.
Xueyao Zhang, Liumeng Xue, Yuancheng Wang +10 authors
Amphion is a toolkit for audio, music, and speech generation that includes model visualizations, vocoders, and evaluation metrics to support reproducible research and training for researchers.
Gemini Team, Rohan Anil, Sebastian Borgeaud +939 authors
Gemini, a family of multimodal models, achieves state-of-the-art performance across various benchmarks, including human-expert performance on MMLU, through advanced cross-modal reasoning and language understanding.
Dan Kondratyuk, Lijun Yu, Xiuye Gu +28 authors
VideoPoet is a language model that synthesizes high-quality video and audio from diverse inputs using a transformer architecture, showcasing superior zero-shot video generation capabilities.
Yixin Song, Zeyu Mi, Haotong Xie +1 authors
PowerInfer, a high-speed LLM inference engine for personal computers, enhances efficiency using hotspot neuron analysis, GPU-CPU hybrid computation, adaptive predictors, and neuron-aware sparse operators, achieving performance close to server-grade GPUs.
Renat Aksitov, Sobhan Miryoosefi, Zonglin Li +10 authors
A ReAct-style LLM agent with iterative training using reinforcement learning and self-distillation improves complex question-answering performance with fewer parameters.
Juan A. Rodriguez, Shubham Agarwal, Issam H. Laradji +4 authors
StarVector integrates CodeLLMs and vision models to generate complex SVGs by aligning visual and code tokens, enhancing visual quality and complexity handling.
Quan Sun, Yufeng Cui, Xiaosong Zhang +8 authors
A large-scale generative multimodal model with 37 billion parameters demonstrates strong few-shot in-context learning and achieves state-of-the-art performance on multimodal tasks through scaling-up and instruction tuning.
Collin Burns, Pavel Izmailov, Jan Hendrik Kirchner +9 authors
Naive finetuning of strong pretrained models with weak supervision can improve performance but is insufficient to match full model capabilities, highlighting challenges for aligning superhuman models.
Chenxu Zhang, Chao Wang, Jianfeng Zhang +7 authors
DREAM-Talk is a two-stage diffusion-based framework generating expressive emotional talking faces with accurate lip-sync from single portrait images.
Miao Hua, Jiawei Liu, Fei Ding +3 authors
DreamTurner uses a novel approach by injecting reference information through subject encoders and self-subject-attention layers to enhance subject-driven image generation, balancing subject learning and model capabilities.
Saurabh Saxena, Junhwa Hur, Charles Herrmann +2 authors
A generic diffusion model with log-scale depth parameterization and FOV conditioning achieves state-of-the-art zero-shot metric depth estimation by handling indoor and outdoor scenes effectively and reducing relative error significantly.
Bichen Wu, Ching-Yao Chuang, Xiaoyan Wang +6 authors
Fairy, an adaptation of image-editing diffusion models for video editing, uses anchor-based cross-frame attention to enhance temporal coherence and efficiency, outperforming existing methods.
Yifeng Ma, Shiwei Zhang, Jiayu Wang +3 authors
DreamTalk enables expressive talking head generation using diffusion models with specialized components for denoising, lip-sync, and style prediction.
Chaoning Zhang, Dongshen Han, Sheng Zheng +3 authors
A new approach improves MobileSAM's efficiency for the SegEvery task by directly generating valid masks, boosting performance and reducing computation time.
Xianfang Zeng, Xin Chen, Zhongqi Qi +6 authors
Paint3D uses a coarse-to-fine generative framework with specialized diffusion models to produce high-resolution, lighting-less UV texture maps for 3D meshes, addressing challenges of incomplete areas and illumination artifacts.
Vikas Thamizharasan, Difan Liu, Shantanu Agarwal +5 authors
A cascaded diffusion model combines raster and vector diffusion models, using a transformer architecture and novel vector representation to generate high-quality, complex vector fonts.
Xiaoyang Wu, Li Jiang, Peng-Shuai Wang +6 authors
Point Transformer V3 improves efficiency and scaling in point cloud processing, achieving state-of-the-art performance by simplifying neighbor search and expanding receptive field.
Kai Nylund, Suchin Gururangan, Noah A. Smith
Time vectors, derived by fine-tuning language models on specific time periods, enhance performance on text from those periods and can be interpolated to improve performance on future periods without further training.
Jiahui Gao, Renjie Pi, Jipeng Zhang +8 authors
G-LLaVA outperforms GPT-4-V on solving geometric problems using an enriched multimodal geometry dataset, Geo170K.
Yiming Zhang, Zhening Xing, Yanhong Zeng +2 authors
PIA enhances personalized T2I models to generate animations by integrating temporal alignment and condition modules for motion control and appearance consistency.
Ye Yuan, Xueting Li, Yangyi Huang +4 authors
A method combining Gaussian splatting with neural implicit fields and SDF-based mesh learning for generating realistic and animatable avatars from text prompts, achieving high-quality appearance and geometry with fast rendering.
Zeyinzi Jiang, Chaojie Mao, Yulin Pan +2 authors
An efficient generative tuning framework, SCEdit, leverages skip connections in U-Net to enhance image generation performance and efficiency, supporting multi-condition inputs with controllable tuning modules.
Youwei Liang, Junfeng He, Gang Li +15 authors
Enriched human feedback is used to improve text-to-image generation through multimodal transformer training, enhancing quality and generalizability across different models.
Hangjie Yuan, Shiwei Zhang, Xiang Wang +7 authors
InstructVideo enhances video generation by reward fine-tuning text-to-video diffusion models with human feedback using repurposed image reward models and efficient segmentation techniques.
Mingsheng Li, Xin Chen, Chi Zhang +5 authors
M3DBench is a comprehensive 3D instruction-following dataset enabling a broad range of multimodal 3D tasks and serves as a new benchmark for assessing model performance in understanding 3D prompts.
Mohammad Samragh, Mehrdad Farajtabar, Sachin Mehta +5 authors
A technique called weight subcloning expedites the training of small transformer models using large pretrained models by reducing embedding dimensions and removing blocks.
Hayk Manukyan, Andranik Sargsyan, Barsegh Atanyan +3 authors
HD-Painter, a training-free approach, enhances high-resolution text-guided image inpainting through Prompt-Aware Introverted Attention and Reweighting Attention Score Guidance, achieving superior 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