Generating human motion from textual descriptions is a challenging task. Existing methods either struggle with physical credibility or are limited by the complexities of physics simulations. In this paper, we present ReinDiffuse that combines reinforcement learning with motion diffusion model to generate physically credible human motions that align with textual descriptions. Our method adapts Motion Diffusion Model to output a parameterized distribution of actions, making them compatible with reinforcement learning paradigms. We employ reinforcement learning with the objective of maximizing physically plausible rewards to optimize motion generation for physical fidelity. Our approach outperforms existing state-of-the-art models on two major datasets, HumanML3D and KIT-ML, achieving significant improvements in physical plausibility and motion quality. Project: https://reindiffuse.github.io/

We introduce HuTuMotion, an innovative approach for generating natural human motions that navigates latent motion diffusion models by leveraging few-shot human feedback. Unlike existing approaches that sample latent variables from a standard normal prior distribution, our method adapts the prior distribution to better suit the characteristics of the data, as indicated by human feedback, thus enhancing the quality of motion generation. Furthermore, our findings reveal that utilizing few-shot feedback can yield performance levels on par with those attained through extensive human feedback. This discovery emphasizes the potential and efficiency of incorporating few-shot human-guided optimization within latent diffusion models for personalized and style-aware human motion generation applications. The experimental results show the significantly superior performance of our method over existing state-of-the-art approaches.

For hyperspectral image, it is a challenging task to select informative and distinctive bands due to the lack of labeled samples and massive redundancy. To address this issue, we propose a new unsupervised band selection method via Sparse Principal Component Analysis and Adaptive Multiple Graph Learning (SPCA-AMGL). Based on PCA, it proposes a Sparse PCA with L 2,1 norm sparse constraint, which can effectively select the bands with high information and low correlation. In addition, an adaptive multiple graph learning is used for manifold-preserving, which ensures that the bands containing abundant spatial structure information are preserved. Specifically, it constructs multiple initial similarity graphs with different distance metrics, and then learns an adaptive graph from them. In this way, it overcomes the shortcoming of insufficient intrinsic structure of data learned from a single graph. Experimental result on Indian Pines data set proves the effectiveness and advancement of SPCA-AMGL. The source code is available at: https://github.com/ZWX0823/SPCA-AMGL.

How to achieve hair editing has always been a longstanding topic. The task of hair editing expects users to create a draft to customize hairstyle, textures and colors. Then it will translate the hair draft to corresponding authentic hair. However, hair generally has complex feature representations due to delicate textures and rich colors, which means it difficult to retain all the hair features during the generation process. Thus, how to reconstruct hair with visually pleasant textures and diverse colors is a challenge. To address this problem, this paper proposes a novel method that takes Image-to-Image Generative Adversarial Networks as the backbone and induces two-phase training mode to capture different hair features. In the first phase, networks take the hair colour draft as input to reconstruct low-frequency information. In the next phase, the result output in the first phase will be fed into the network again to refine hair. Our extensive experiments show that two-phase training mode can build up convincing hair images and also allows users to customize the style of hair in the textures as well as colors.