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
Kai Zou, Dian Zheng, Hongbo Liu +3 authors
HiAR, a hierarchical autoregressive diffusion framework, improves video generation by conditioning on context at the same noise level, enabling faster inference and better temporal consistency.
Yinpei Dai, Hongze Fu, Jayjun Lee +6 authors
RoboMME presents a large-scale standardized benchmark for evaluating vision-language-action models in long-horizon robotic manipulation tasks requiring memory mechanisms.
Kangsan Kim, Yanlai Yang, Suji Kim +4 authors
Multi-agent systems require understanding multiple long-horizon egocentric videos simultaneously, necessitating new benchmarks and models for system-level comprehension.
Kaiser Sun, Xiaochuang Yuan, Hongjun Liu +4 authors
Multimodal large language models exhibit inconsistent performance when processing text from images versus textual tokens, with factors like rendering quality and task type influencing this modality gap, which can be mitigated through self-distillation techniques that leverage text-based reasoning traces.
Hongfei Zhang, Harold Haodong Chen, Chenfei Liao +12 authors
Video depth estimation framework DVD adapts pre-trained video diffusion models into deterministic single-pass depth regressors using structural anchors, latent manifold rectification, and global affine coherence for improved accuracy and efficiency.
Yuchen Yang, Yuqing Shao, Duxiu Huang +11 authors
CourtSI is a large-scale spatial intelligence dataset for sports scenarios that enables evaluation and improvement of vision-language models' understanding of human motion and object interactions.
Qianyu Yang, Yang Liu, Jiaqi Li +20 authors
A new benchmark evaluates language models on complex, real-world professional tasks requiring multi-step reasoning, evidence resolution, and domain-specific decision-making across multiple industries.
Ruizhong Qiu, Hanqing Zeng, Yinglong Xia +15 authors
Researchers address imbalance in routing weights of Mixture-of-LoRAs models by proposing Reinforcement Routing (ReMix), which uses non-learnable weights and reinforcement learning techniques to improve model expressiveness and performance.
Hui Zhang, Juntao Liu, Zongkai Liu +4 authors
WeEdit presents a systematic approach for text-centric image editing with a scalable data pipeline, multi-language benchmarks, and a two-stage training strategy combining supervised fine-tuning and reinforcement learning.
Aviad Dahan, Moran Yanuka, Noa Kraicer +2 authors
ID-LoRA enables joint generation of visual appearance and voice using a single model, improving personalization through identity-driven in-context LoRA adaptation and identity guidance techniques.
Xiangyu Zhao, Peiyuan Zhang, Junming Lin +7 authors
Reinforcement learning framework with novel reward modeling and benchmarking approaches improves fidelity and instruction adherence in image editing and text-to-image generation.
Avihu Dekel, Samuel Thomas, Takashi Fukada +1 authors
A non-autoregressive speech recognition approach formulates acoustic-to-text conversion as conditional transcript editing using a bidirectional language model editor with latent alignment training and interleaved padding for improved efficiency.
Nilesh Jain, Rohit Yadav, Sagar Kotian +1 authors
An autonomous reinforcement learning framework conducts continuous neural architecture and hyperparameter research without human intervention, achieving performance comparable to hand-tuned baselines through automated experimentation.
Moayed Haji-Ali, Willi Menapace, Ivan Skorokhodov +6 authors
Elastic Latent Interface Transformer (ELIT) decouples compute from image resolution in diffusion transformers by introducing learnable latent tokens that adaptively prioritize important regions, enabling dynamic resource allocation without modifying core model architecture.
Yi-Kai Zhang, Yueqing Sun, Hongyan Hao +4 authors
Adaptive value estimation method combines pretrained prior with empirical rollouts using real-time statistical testing to reduce variance and improve reinforcement learning performance under sparse sampling conditions.
Tzu-Heng Huang, Sirajul Salekin, Javier Movellan +2 authors
RubiCap introduces a reinforcement learning framework for dense image captioning that uses large language model-generated rubrics to provide fine-grained reward signals, achieving superior performance over supervised methods and competing with larger models.
Karan Gupta, Pranav Vajreshwari, Yash Pandya +3 authors
ATLAS enables small language models to effectively operate in large-scale tool environments through reinforcement fine-tuning that learns context control and execution structure, achieving performance comparable to larger models under stricter resource constraints.
Zuhao Zhang, Chengyue Yu, Yuante Li +3 authors
MiniAppBench introduces the first comprehensive benchmark for evaluating principle-driven, interactive application generation, addressing the gap in existing benchmarks that focus on static correctness rather than dynamic, real-world interactions.
Soumik Mukhopadhyay, Prateksha Udhayanan, Abhinav Shrivastava
Scale-space theory connects diffusion models' information hierarchy to low-pass filtering, leading to a framework that combines scale spaces with diffusion processes for efficient image processing.
Akash Haridas, Utkarsh Saxena, Parsa Ashrafi Fashi +3 authors
Dynamic Chunking Diffusion Transformer adapts token sequence length based on image content and diffusion timestep, improving efficiency and performance over fixed-token approaches.
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