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。
Le Xue, Manli Shu, Anas Awadalla +24 authors
xGen-MM, an extension of Salesforce's xGen initiative, provides a framework for developing Large Multimodal Models with pre-trained, instruction-tuned, and safety-tuned variants.
50 篇论文 · 按点赞排序
Rawal Khirodkar, Timur Bagautdinov, Julieta Martinez +5 authors
Sapiens, a family of vision models for human-centric tasks, achieves superior performance with self-supervised pretraining and can be easily fine-tuned for 2D pose estimation, body-part segmentation, depth estimation, and surface normal prediction.
Xun Liang, Hanyu Wang, Yezhaohui Wang +8 authors
Controllable Text Generation techniques for Large Language Models ensure predefined control conditions and high-quality text output, covering content and attribute control through various methods like retraining, fine-tuning, and latent manipulation.
Qianqian Xie, Dong Li, Mengxi Xiao +36 authors
Open-FinLLMs, a series of financial LLMs, enhance performance through comprehensive pre-training and instruction fine-tuning, demonstrating superior results in both zero-shot and few-shot evaluations, and excels in understanding tables and charts.
Chunting Zhou, Lili Yu, Arun Babu +7 authors
Transfusion combines language modeling and diffusion training to develop a multi-modal transformer model that scales effectively across text and image data, matching the performance of specialized models.
Sharath Turuvekere Sreenivas, Saurav Muralidharan, Raviraj Joshi +6 authors
Models Llama 3.1 8B and Mistral NeMo 12B are compressed using pruning and distillation, achieving state-of-the-art performance in a reduced parameter setting.
Ryan Teknium, Jeffrey Quesnelle, Chen Guang
Hermes 3, a neutrally-aligned instruct and tool use model with strong reasoning and creative capabilities, achieves top performance on public benchmarks.
Hyeongmin Lee, Jin-Young Kim, Kyungjune Baek +18 authors
The study introduces TWLV-I, a novel video foundation model that outperforms existing models in video comprehension by constructing robust representations for both motion and appearance.
Xianjie Wu, Jian Yang, Linzheng Chai +10 authors
TableLLM, trained on TableInstruct, benchmarks performance in tabular data reasoning and highlights the need for improvement in real-world applications.
Fuzhao Xue, Yukang Chen, Dacheng Li +15 authors
LongVILA, a full-stack solution for long-context vision-language models, introduces Multi-Modal Sequence Parallelism for efficient training and inference, and a five-stage training pipeline that enhances long video processing and captioning performance.
Jinheng Xie, Weijia Mao, Zechen Bai +7 authors
Show-o, a unified transformer model, combines autoregressive and diffusion modeling for versatile vision-language tasks, achieving competitive performance.
Viraat Aryabumi, Yixuan Su, Raymond Ma +6 authors
The inclusion of code data during pre-training significantly improves LLMs' performance across various natural language and world knowledge tasks, beyond code generation.
Xiaochuang Han, Marjan Ghazvininejad, Pang Wei Koh +1 authors
Using canonical codec representations (JPEG, AVC) as a direct modeling approach for image and video generation outperforms pixel-based and vector quantization methods, facilitating multi-modal LLMs.
Shengran Hu, Cong Lu, Jeff Clune
A new research area, Automated Design of Agentic Systems (ADAS), uses a meta agent to automatically create powerful agentic system designs, demonstrating superior performance and robustness across various domains.
Can Qin, Congying Xia, Krithika Ramakrishnan +16 authors
The xGen-VideoSyn-1 model generates realistic videos from text using a Latent Diffusion Model and Video Variational Autoencoder, with a spatial and temporal divide-and-merge strategy and Diffusion Transformer.
Minghua Liu, Chong Zeng, Xinyue Wei +9 authors
MeshFormer, a sparse-view 3D reconstruction model, enhances geometry learning and mesh generation by integrating 3D sparse voxels, transformers, 3D convolutions, 2D diffusion models, SDF supervision, and surface rendering.
Jamba Team, Barak Lenz, Alan Arazi +58 authors
Jamba-1.5, a hybrid Transformer-Mamba model, offers high throughput and low memory usage while achieving high quality in instruction-following tasks, supported by ExpertsInt8 quantization.
Weiliang Chen, Fangfu Liu, Diankun Wu +3 authors
DreamCinema uses generative AI to streamline the creation of high-quality films with user-friendly 3D characters and smooth camera movements, integrating cinematic transfer and motion generation.
Khang T. Doan, Bao G. Huynh, Dung T. Hoang +5 authors
Vintern-1B, a multimodal large language model, integrates Qwen2 and InternViT for OCR, document extraction, and Vietnamese question-answering by fine-tuning on over 3 million image-question-answer pairs.
Artem Snegirev, Maria Tikhonova, Anna Maksimova +2 authors
A new Russian-focused embedding model, ru-en-RoSBERTa, and a Russian benchmark, ruMTEB, are introduced, demonstrating competitive performance in various NLP tasks compared to existing state-of-the-art models.
Haopeng Li, Jinyue Yang, Kexin Wang +4 authors
AiM, an autoregressive image generative model using the Mamba state-space architecture, achieves superior quality and faster inference speed compared to existing AR and diffusion models.
Zhenyu Li, Yike Zhang, Tengyu Pan +6 authors
FocusLLM extends the context length of decoder-only LLMs efficiently by chunking long texts and using a parallel decoding mechanism to enhance performance on long-context tasks.
Aoran Xiao, Weihao Xuan, Heli Qi +3 authors
MM-SAM extends SAM to support multi-modal and cross-modal processing, enhancing segmentation accuracy across various sensor modalities through unsupervised cross-modal transfer and weakly-supervised multi-modal fusion.
Haofeng Liu, Erli Zhang, Junde Wu +2 authors
SurgSAM-2 enhances surgical video segmentation using an Efficient Frame Pruning mechanism, improving real-time performance and accuracy compared to SAM2.
Zongze Wu, Nicholas Kolkin, Jonathan Brandt +2 authors
The paper presents an iterative inversion technique using few-step diffusion models, enabling precise image inversion and disentangled editing through text prompts, achieving real-time edits with low functional evaluations and superior performance over existing methods.
Haitao Zhou, Chuang Wang, Rui Nie +4 authors
TrackGo, using free-form masks and arrows along with TrackAdapter, achieves top performance in controlled video generation by enhancing pretrained temporal self-attention layers.
Xuanlei Zhao, Xiaolong Jin, Kai Wang +1 authors
Pyramid Attention Broadcast (PAB) enhances DiT-based video generation by broadcasting attention outputs with a pyramid structure, achieving real-time performance and efficiency improvements.
Dimitrios Tsaras, Antoine Grosnit, Lei Chen +3 authors
ShortCircuit, a transformer-based architecture, uses supervised and reinforcement learning to generate optimal AIGs from truth tables, outperforming traditional methods in circuit size.
Tao Yang, Yangming Shi, Yunwen Huang +3 authors
A factorized spatiotemporal framework, Factorized-Dreamer, enables high-quality video generation from limited and low-quality video data and concise captions.
Zhiyong Zhang, Aniket Gupta, Huaizu Jiang +1 authors
A new optical flow method based on NeuFlow v1 achieves high accuracy with significantly reduced computational cost, offering a 10x-70x speedup over state-of-the-art methods.
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