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。
Jianwen Jiang, Chao Liang, Jiaqi Yang +3 authors
An audio-only conditioned video diffusion model, Loopy, improves natural and high-quality human video generation without auxiliary spatial signals.
50 篇论文 · 按点赞排序
Vadim Titov, Madina Khalmatova, Alexandra Ivanova +2 authors
A modified diffusion sampling process using self-guidance and noise rescaling achieves high-quality image editing with preserved structure and appearance without requiring diffusion model fine-tuning.
Zifan Zheng, Yezhaohui Wang, Yuxin Huang +4 authors
This survey explores the interpretability and underlying mechanisms of attention heads within Large Language Models to understand their reasoning processes.
Haoran Wei, Chenglong Liu, Jinyue Chen +9 authors
GOT, a unified OCR model with 580M parameters, handles various types of characters and OCR tasks and offers advanced features like region-level recognition and dynamic resolution.
Niklas Muennighoff, Luca Soldaini, Dirk Groeneveld +21 authors
A sparse Mixture-of-Experts language model with 7 billion parameters achieves superior performance by using only 1 billion parameters per input token and outperforms larger models in various experiments.
Sushant Gautam, Andrea Storås, Cise Midoglu +4 authors
Kvasir-VQA is a dataset with question-and-answer annotations for GI diagnostics, supporting image captioning, VQA, synthetic image generation, object detection, and classification.
Xidong Wang, Dingjie Song, Shunian Chen +2 authors
LongLLaVA, a hybrid MLLM combining Mamba and Transformer blocks, enhances video and image understanding with improved efficiency and throughput.
Zhifei Xie, Changqiao Wu
Mini-Omni is an end-to-end audio-based conversational model enabling real-time speech interaction and efficient speech generation.
jiajie Zhang, Yushi Bai, Xin Lv +8 authors
A new pipeline and dataset enable long-context LLMs to generate responses with precise sentence-level citations, improving their trustworthiness and outperforming existing models.
Liqun Yang, Jian Yang, Chaoren Wei +13 authors
A framework uses fine-tuned large language models to guide input mutation in fuzzing, improving the detection of software vulnerabilities.
Zhiyuan Hu, Yuliang Liu, Jinman Zhao +8 authors
LongRecipe extends the context window of large language models efficiently with reduced resources and preserved performance, achieving a 16-fold increase from 8k to 128k context size.
Mouxiang Chen, Lefei Shen, Zhuo Li +3 authors
The VisionTS model, leveraging a visual masked autoencoder pre-trained on ImageNet, achieves state-of-the-art zero-shot and fine-tuned forecasting performance in time series forecasting without adapting specifically to the time-series domain.
Wenbo Hu, Xiangjun Gao, Xiaoyu Li +5 authors
DepthCrafter generates temporally consistent depth sequences for open-world videos using a video-to-depth model derived from a pre-trained image-to-video diffusion model.
Sihang Li, Jin Huang, Jiaxi Zhuang +7 authors
A hybrid approach combining continual pre-training and supervised fine-tuning creates specialized LLMs for scientific literature, addressing challenges in constructing corpora and generating instructions.
Xiang Yue, Tianyu Zheng, Yuansheng Ni +11 authors
MMMU-Pro is a benchmark that enhances the evaluation of multimodal models by filtering text-only answerable questions, augmenting options, and introducing vision-only inputs, demonstrating challenges in integrated visual and textual understanding.
Songhua Liu, Weihao Yu, Zhenxiong Tan +1 authors
A novel linear attention mechanism, named LinFusion, is introduced to enhance high-resolution image generation with reduced computational cost, offering performance comparable to or better than existing diffusion models.
Zhengcong Fei, Mingyuan Fan, Changqian Yu +1 authors
FluxMusic extends diffusion-based rectified flow Transformers to generate music from text by transforming it into a latent VAE space of mel-spectra and using text encoders and modulations for enhanced semantic capture.
Jifan Yu, Zheyuan Zhang, Daniel Zhang-li +30 authors
MAIC, leveraging large language models and multi-agent systems, aims to improve online education through scalability and adaptivity, supported by a collaborative platform for research and innovation.
Yuxuan Wang, Cihang Xie, Yang Liu +1 authors
VideoLLaMB enhances video-language model performance through temporal memory tokens and SceneTilling algorithm, enabling accurate frame retrieval and efficient video caption generation without additional training.
Slava Elizarov, Ciara Rowles, Simon Donné
Geometry Image Diffusion leverages 2D representation and collaborative control from Text-to-Image models to generate high-quality 3D objects efficiently.
Anwen Hu, Haiyang Xu, Liang Zhang +6 authors
A high-resolution compression module and three-stage training framework enhance multi-page document understanding and reduce latency while maintaining performance.
Baichuan Zhou, Haote Yang, Dairong Chen +7 authors
UrBench is a comprehensive benchmark for evaluating Large Multimodal Models in complex, multi-view urban scenarios, revealing gaps in model performance compared to humans.
Feize Wu, Yun Pang, Junyi Zhang +5 authors
Context Regularization, a method for embedding new concepts in text encoding, enhances text-to-image personalization by improving semantic understanding and generalization, leading to better identity preservation and text alignment.
Fan Zhang, Michael Gienger
A framework for assistive robot manipulation uses parameter-efficient prompt tuning and supervised Flow Matching to learn manipulation affordances and robot trajectories in daily living scenarios.
Allen Z. Ren, Justin Lidard, Lars L. Ankile +6 authors
DPPO, a policy gradient method for fine-tuning diffusion-based policies, outperforms other RL methods in continuous control and robot learning tasks, offering stable training, robust policies, and effective zero-shot deployment.
Bin Wang, Fan Wu, Linke Ouyang +5 authors
A new Character Detection Matching (CDM) metric is proposed for formula recognition evaluation, offering greater accuracy and fairness by considering spatial and character-level matching of image-formatted formulas.
Yuntian Deng, Wenting Zhao, Jack Hessel +3 authors
WildVis is an interactive tool for large-scale conversation analysis, providing search and visualization capabilities in text and embedding spaces with optimizations for handling million-scale datasets.
Yuxiang Wei, Hojae Han, Rajhans Samdani
Arctic-SnowCoder-1.3B, a data-efficient code model pretrained on progressively refined high-quality data, outperforms larger models on coding benchmarks by aligning data with downstream applications.
Wei Xiong, Chengshuai Shi, Jiaming Shen +10 authors
A multi-turn direct preference learning framework, incorporating feedback from code interpreters, enhances large language models' mathematical problem-solving capabilities with substantial performance improvements.
Hanxin Zhu, Tianyu He, Anni Tang +3 authors
A novel text-to-video generation method decomposes concepts into 3D representations using LLMs, composes them with multi-modal LLM guidance, and refines with 2D diffusion models to produce high-fidelity videos with controlled motion and flexibility.
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