Recently, few-shot learning-based methods have achieved impressive results in cross-domain hyperspectral classification. However, existing approaches often ignore differences in spectral information caused by varing spectral range across different datasets, and encounter limitations due to the constrained parameter size of the models. Furthermore, the substantial differences between RGB and hyperspectral images present significant challenges in applying foundation models (e.g., SAM, DINOv2) to the hyperspectral domain. This paper proposes a novel framework for cross-domain few-shot hyperspectral classification that leverages parameter-efficient fine-tuning, which we apply to DINOv2 to construct SpectralDINO. Different spectral ranges reflect different physical properties of the target. To enhance the consistency of the spectral features extracted by the model from different domains, we introduce a source domain spectral alignment (SDSA) strategy to align the spectral ranges and bands of the source domain data to the target domain. SpectralDINO employs the visual foundation model to enhance its ability to generalize cross-domain knowledge. Additionally, we propose a dual mixture-of-subspaces low-rank adaptation (Dual-MoS LoRA) method to address the structural limitation of the low-rank adaptation methods in distinguishing domain-specific features from multi-domain inputs. Only 1.14% of the 21.37M parameters need to be trained to perform fine-tuning. Extensive experimental results on three public datasets demonstrate the superiority of SpectralDINO.

We propose a novel model-based reinforcement learning algorithm---Dynamics Learning and predictive control with Parameterized Actions (DLPA)---for Parameterized Action Markov Decision Processes (PAMDPs). The agent learns a parameterized-action-conditioned dynamics model and plans with a modified Model Predictive Path Integral control. We theoretically quantify the difference between the generated trajectory and the optimal trajectory during planning in terms of the value they achieved through the lens of Lipschitz Continuity. Our empirical results on several standard benchmarks show that our algorithm achieves superior sample efficiency and asymptotic performance than state-of-the-art PAMDP methods.

We design a system that learns how to edit visual programs. Our edit network consumes a complete input program and a visual target. From this input, we task our network with predicting a local edit operation that could be applied to the input program to improve its similarity to the target. In order to apply this scheme for domains that lack program annotations, we develop a self-supervised learning approach that integrates this edit network into a bootstrapped finetuning loop along with a network that predicts entire programs in one-shot. Our joint finetuning scheme, when coupled with an inference procedure that initializes a population from the one-shot model and evolves members of this population with the edit network, helps to infer more accurate visual programs. Over multiple domains, we experimentally compare our method against the alternative of using only the one-shot model, and find that even under equal search-time budgets, our editing-based paradigm provides significant advantages.

Temporal Action Segmentation (TAS) has achieved great success in many fields such as exercise rehabilitation, movie editing, etc. Currently, task-driven TAS is a central topic in human action analysis. However, motion-centered TAS, as an important topic, is little researched due to unavailable datasets. In order to explore more models and practical applications of motion-centered TAS, we introduce a Motion-Centered Figure Skating (MCFS) dataset in this paper. Compared with existing temporal action segmentation datasets, the MCFS dataset is fine-grained semantic, specialized and motion-centered. Besides, RGB-based and Skeleton-based features are provided in the MCFS dataset. Experimental results show that existing state-of-the-art methods are difficult to achieve excellent segmentation results (including accuracy, edit and F1 score) in the MCFS dataset. This indicates that MCFS is a challenging dataset for motion-centered TAS. The latest dataset can be downloaded at https://shenglanliu.github.io/mcfs-dataset/.