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。
Ruihang Chu, Yefei He, Zhekai Chen +10 authors
Wan-Move enhances motion control in video generative models by integrating motion-aware features into latent space, enabling high-quality and scalable video synthesis.
50 篇论文 · 按点赞排序
Dmitrii Stoianov, Danil Taranets, Olga Tsymboi +12 authors
T-pro 2.0 is an open-weight Russian LLM for hybrid reasoning and efficient inference, using a Cyrillic-dense tokenizer and EAGLE speculative-decoding pipeline.
Tong Wu, Yang Liu, Jun Bai +6 authors
NPR, a teacher-free framework, enhances Large Language Models with native parallel reasoning capabilities through self-distilled training, Parallel-Aware Policy Optimization, and a robust NPR Engine, achieving substantial performance and speed improvements.
Yuning Gong, Yifei Liu, Yifan Zhan +21 authors
Visionary is an open web-native platform enabling real-time rendering of 3D Gaussian Splatting and meshes with efficient GPU-based inference, supporting dynamic content and generative models.
Zhenglin Cheng, Peng Sun, Jianguo Li +1 authors
TwinFlow is a 1-step generative model framework that enhances inference efficiency without requiring fixed pretrained teacher models or standard adversarial networks, achieving high performance on text-to-image tasks and scaling efficiently.
Ke Xing, Longfei Li, Yuyang Yin +8 authors
StereoWorld generates high-quality stereo video from monocular input using a pretrained video generator with geometry-aware regularization and spatio-temporal tiling.
Xiaoran Liu, Yuerong Song, Zhigeng Liu +6 authors
The paper proposes a method to enhance Rotary Position Embeddings by utilizing both the real and imaginary components of the complex-valued dot product, improving long-context modeling in Large Language Models.
Cheng Liu, Yiren Song, Haofan Wang +1 authors
OmniPSD, a diffusion framework within the Flux ecosystem, enables text-to-PSD generation and image-to-PSD decomposition, achieving high-fidelity results with transparency awareness.
Zekai Luo, Zongze Du, Zhouhang Zhu +7 authors
LivingSwap enhances video face swapping by using keyframes and reference guidance to maintain identity and fidelity over long sequences, reducing manual effort and achieving state-of-the-art results.
Xiangpeng Yang, Ji Xie, Yiyuan Yang +3 authors
VideoCoF, a Chain-of-Frames approach, improves video editing precision and instruction-to-region mapping by using reasoning tokens without requiring user-provided masks.
Yiwen Tang, Zoey Guo, Kaixin Zhu +11 authors
This study investigates reinforcement learning for text-to-3D generation, focusing on reward designs, RL algorithms, benchmarking, and hierarchical optimization, introducing AR3D-R1 as the first RL-enhanced model for 3D generation.
Songyang Gao, Yuzhe Gu, Zijian Wu +18 authors
OPV, an iterative active learning framework with Rejection Fine-Tuning, enhances verification of long reasoning chains in large language models, achieving state-of-the-art results and improving accuracy in collaborative tasks.
Zhaochong An, Menglin Jia, Haonan Qiu +12 authors
OneStory generates coherent multi-shot videos by modeling global cross-shot context through a Frame Selection module and an Adaptive Conditioner, leveraging pretrained image-to-video models and a curated dataset.
Yi-Chuan Huang, Jiewen Chan, Hao-Jen Chien +1 authors
Voxify3D is a two-stage framework that combines 3D mesh optimization with 2D pixel art supervision to generate high-quality voxel art with semantic preservation, pixel-art aesthetics, and discrete color coherence.
Charlie Zhang, Graham Neubig, Xiang Yue
A controlled experimental framework isolates and evaluates the contributions of pre-training, mid-training, and reinforcement learning in improving language model reasoning, demonstrating the necessity of each phase and the role of process-level rewards.
Navve Wasserman, Matias Cosarinsky, Yuval Golbari +4 authors
An automated framework uses unsupervised decomposition and natural language descriptions to identify and explain visual representations in human brain fMRI data.
Hongyu Li, Manyuan Zhang, Dian Zheng +11 authors
A deliberative editing framework with a reasoning engine improves instruction-following in image editing through iterative critique and refinement, significantly enhancing performance.
Zijian Wu, Lingkai Kong, Wenwei Zhang +12 authors
The Outcome-based Process Verifier (OPV) improves the verification of complex reasoning chains in large language models by combining outcome-based and process-based verification with iterative active learning and Rejection Fine-Tuning, achieving state-of-the-art performance on various benchmarks.
Tarun Suresh, Nalin Wadhwa, Debangshu Banerjee +1 authors
BEAVER is a framework that provides deterministic and sound probability bounds for verifying constraints in large language models, achieving tighter bounds and identifying more high-risk instances than baseline methods.
Zongwei Li, Zhonghang Li, Zirui Guo +2 authors
DeepCode, a fully autonomous framework, addresses the challenges of document-to-codebase synthesis by optimizing information flow through source compression, structured indexing, knowledge injection, and error correction, achieving state-of-the-art performance and surpassing human experts.
Haiteng Zhao, Junhao Shen, Yiming Zhang +7 authors
InternGeometry, an LLM agent, surpasses human performance on IMO geometry problems using a heuristic-driven approach with iterative proposition verification and a dynamic memory mechanism, significantly outperforming AlphaGeometry 2 with limited training data.
Kehong Gong, Zhengyu Wen, Weixia He +8 authors
MoCapAnything is a reference-guided framework that reconstructs rotation-based animations from monocular video for arbitrary rigged 3D assets, enabling cross-species retargeting and scalable 3D motion capture.
Sen Ye, Jianning Pei, Mengde Xu +4 authors
DMVAE explicitly aligns the encoder's latent distribution with a reference distribution, improving modeling efficiency and image synthesis fidelity compared to conventional VAEs.
Zijian Zhou, Shikun Liu, Haozhe Liu +14 authors
Saber is a scalable zero-shot framework for reference-to-video generation that uses video-text pairs to learn identity-consistent representations and outperforms models trained with explicit reference data.
Runjia Li, Moayed Haji-Ali, Ashkan Mirzaei +9 authors
EgoEdit is a real-time, instruction-following egocentric video editor that addresses challenges in handling egomotion and hand-object interactions, outperforming existing methods on egocentric editing tasks.
Xianghao Kong, Zeyu Zhang, Yuwei Guo +3 authors
Bind & Compose uses Diffusion Transformers with hierarchical binders and temporal strategies to accurately compose complex visual concepts from images and videos.
Ming Ma, Jue Zhang, Fangkai Yang +4 authors
DoVer, an intervention-driven debugging framework, enhances reliability in LLM-based multi-agent systems by actively validating failure hypotheses and measuring task progress through targeted interventions.
Changpeng Yang, Jinyang Wu, Yuchen Liu +9 authors
CAPO, a curriculum advantage policy optimization, enhances reinforcement learning for large language models by strategically introducing positive and negative advantage signals, improving reasoning capabilities and generalization.
Junyan Ye, Leiqi Zhu, Yuncheng Guo +7 authors
RealGen is a photorealistic text-to-image framework that uses an LLM for prompt optimization and a diffusion model for image generation, enhanced by a Detector Reward mechanism and RealBench for automated evaluation.
Yixin Wan, Lei Ke, Wenhao Yu +2 authors
MotionEdit is a new dataset for motion-centric image editing, and MotionNFT is a post-training framework that enhances motion editing quality by aligning motion flow between input and edited images.
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