MUMINS: Metadata-conditioned Uncertainty-aware Medical Image Next-state Synthesis
本文提出MUMINS,一种基于元数据条件和不确定性感知的医学图像未来状态合成方法,有效预测解剖学变化并量化不确定性。
本文提出MUMINS,一种基于元数据条件和不确定性感知的医学图像未来状态合成方法,有效预测解剖学变化并量化不确定性。
本文通过结合面部几何特征、类别自适应模态融合和情感几何先验的方法,提高了对话中多模态情感识别的准确性。
This work addresses the challenge of efficiently adapting general-purpose imitation policies to novel task objectives and constraints while maintaining data efficiency and deployment robustness. The authors propose an instruction-conditioned policy optimization framework that integrates imitation learning with reinforcement learning, leveraging natural language task descriptions to automatically generate reward functions. For the first time, this approach combines human feedback on intermediate trajectories with a Eureka-style reward generation mechanism to enable personalized policy refinement. Evaluated on simulated pick-and-place tasks, the method significantly outperforms feedback-free baselines, achieving enhanced robustness with reduced computational overhead and enabling efficient reuse of general policies across diverse task configurations.
This work addresses the challenging problem of 3D human body reconstruction from sparse, uncalibrated multi-view RGB images depicting clothed subjects. Methodologically, we propose a retraining-free multi-view joint optimization framework built upon the SMPL-X parametric model. We design a cross-view consistent body optimization algorithm and introduce a front-back normal map integration mechanism to explicitly capture geometric details such as clothing wrinkles and hairstyles. Surface fidelity is further enhanced via multi-view joint fitting coupled with normal-guided geometric refinement. Experiments demonstrate that our method achieves superior reconstruction quality compared to single-view baselines—using only 2–4 uncalibrated views—and attains state-of-the-art performance among few-shot 3D human reconstruction approaches. The framework is both computationally efficient and highly generalizable across diverse clothing and pose configurations.
本文提出MUMINS,一种基于元数据条件和不确定性感知的医学图像未来状态合成方法,有效预测解剖学变化并量化不确定性。
本文通过结合面部几何特征、类别自适应模态融合和情感几何先验的方法,提高了对话中多模态情感识别的准确性。
This work addresses the challenge of efficiently adapting general-purpose imitation policies to novel task objectives and constraints while maintaining data efficiency and deployment robustness. The authors propose an instruction-conditioned policy optimization framework that integrates imitation learning with reinforcement learning, leveraging natural language task descriptions to automatically generate reward functions. For the first time, this approach combines human feedback on intermediate trajectories with a Eureka-style reward generation mechanism to enable personalized policy refinement. Evaluated on simulated pick-and-place tasks, the method significantly outperforms feedback-free baselines, achieving enhanced robustness with reduced computational overhead and enabling efficient reuse of general policies across diverse task configurations.
This work addresses the challenging problem of 3D human body reconstruction from sparse, uncalibrated multi-view RGB images depicting clothed subjects. Methodologically, we propose a retraining-free multi-view joint optimization framework built upon the SMPL-X parametric model. We design a cross-view consistent body optimization algorithm and introduce a front-back normal map integration mechanism to explicitly capture geometric details such as clothing wrinkles and hairstyles. Surface fidelity is further enhanced via multi-view joint fitting coupled with normal-guided geometric refinement. Experiments demonstrate that our method achieves superior reconstruction quality compared to single-view baselines—using only 2–4 uncalibrated views—and attains state-of-the-art performance among few-shot 3D human reconstruction approaches. The framework is both computationally efficient and highly generalizable across diverse clothing and pose configurations.