Foundation segmentation models, such as SAM and its video-oriented variant SAM2, have achieved remarkable success in natural image and video segmentation. However, their direct application to echocardiography video is challenged by structural uncertainty arising from severe speckle noise and blurry anatomical boundaries. To address this, we propose E³SAM2, a lightweight adaptation framework that introduces a novel entropy-based methodology to explicitly model and mitigate such uncertainty. Specifically, an entropy-guided attention mechanism is introduced to steer the model’s focus toward structurally reliable features, particularly in speckle-dominated regions. Additionally, an entropy regularization loss is introduced to further enhance target-background discrimination. To better resolve indistinct anatomical contours, an edge-aware supervision module is incorporated to inject explicit boundary priors for sharper delineation. These components are efficiently integrated through a global-local feature adapter. Experiments on CAMUS and EchoNet-Dynamic datasets demonstrate that E³SAM2 achieves state-of-the-art segmentation and clinical estimation performance, while maintaining high computational efficiency.

Deep neural networks (DNNs) have demonstrated outstanding performance in remote sensing imagery (RSI) analysis. However, their sensitivity to adversarial perturbations remains a critical bottleneck for their deployment in security-sensitive scenarios. Currently, most mainstream adversarial purification methods focus on pixel space example reconstruction, neglecting the effective recovery of deep semantic features and the classification decision space, thereby failing to mitigate the adversarial attacks adequately. To address this challenge, we propose a novel multimodal adversarial purification framework, Leveraging Learnable Defense Prompts and a one-step diffusion model (LDP-Diff), which aims to achieve robust adversarial defense with semantic consistency. Firstly, we introduce a learnable defense prompt (LDP) mechanism in LDP-Diff. Through the learnable optimization of input prompts, LDP-Diff can provide precise semantic guidance for the purification process. Next, we introduce a One-Step Diffusion Model combined with the low-rank adaptation (LoRA) for lightweight fine-tuning, significantly reducing the inference complexity while effectively removing adversarial perturbations and achieving high-quality structural and semantic reconstruction. Moreover, to further enhance the stability of purified examples in the classification space, we propose a entropy-guided logit fusion (EGLF) loss, which quantifies the confidence entropy and adaptively fuses the logit distributions between examples to alleviate the class bias remaining in the classification space after reconstruction. Comprehensive experimental results show that LDP-Diff exhibits excellent defense capabilities in multiple RSI datasets. LDP-Diff achieves an accuracy improvement of 2.40% on the UCM dataset and realizes a significant performance improvement of 10.84% on the AID dataset compared to the existing state-of-the-art (SOTA) methods. Notably, LDP-Diff exhibits excellent robustness, generalization, computational efficiency, and significant advantages in cross-domain defense.

This paper proposes a decoder-side Cross Resolution Synthesis (CRS) module to pursue better compression efficiency beyond the latest Versatile Video Coding (VVC), where we encode intra frames at original high resolution (HR), compress inter frames at a lower resolution (LR), and then super-resolve decoded LR inter frames with the help from preceding HR intra and neighboring LR inter frames. For a LR inter frame, a motion alignment and aggregation network (MAN) is devised to produce temporally aggregated motion representation to best guarantee the temporal smoothness; Another texture compensation network (TCN) is utilized to generate texture representation from decoded HR intra frame for better augmenting spatial details; Finally, a similarity-driven fusion engine synthesizes motion and texture representations to upscale LR inter frames for the removal of compression and resolution re-sampling noises. We enhance the VVC using proposed CRS, showing averaged 8.76% and 11.93% Bjntegaard Delta Rate (BD-Rate) gains against the latest VVC anchor in Random Access (RA) and Low-delay P (LDP) settings respectively. In addition, experimental comparisons to the state-of-the-art super-resolution (SR) based VVC enhancement methods, and ablation studies are conducted to further report superior efficiency and generalization of the proposed algorithm. All materials will be made to public at https://njuvision.github.io/CRS for reproducible research.

Summary Machine learning (ML) has been widely applied in many areas in recent decades. However, because of its inherent complexity and characteristics, the efficiency and effectiveness of ML algorithm often to be heavily relies on the technical experts' experience and expertise which play a crucial role to optimize hyperparameters of algorithms. Generally, the procedure tuning the exposed hyperparameters of ML algorithm to achieve better performance is called Hyperparameters Optimization. Traditional hyperparameters optimization methods are manually exhaustive search, which is unavailable for high dimensional search spaces and large datasets. Recent automated sequential model‐based optimization led to substantial improvements for this problem, whose core idea is fitting a regression model to describe the importance and dependence of algorithm's performance on certain given hyperparameter setting. Sequential model‐based algorithm configuration (SMAC) is a the‐state‐of‐art approach, which is specified by four components, Initialize, FitModel, SelectConfigurations, and Intensify. In this article, we propose to add a pruning procedure into SMAC approach, it quantifies the importance of hyperparameters by analyzing the performance of a list of promising configurations and reduces search space by discarding noncritical and bad key hyperparameters. To investigate the impact of pruning for model's performance, we conducted experiments on the configuration space constructed by Auto‐Sklearn and compared the effect of run time and pruning ratio with our algorithm. The experiments results verified that, our method made the configuration selected by SMAC more stable and achieved better performance.

Auscultation of the heart is a widely studied technique, which requires precise hearing from practitioners as a means of distinguishing subtle differences in heart-beat rhythm. This technique is popular due to its non-invasive nature, and can be an early diagnosis aid for a range of cardiac conditions. Machine listening approaches can support this process, monitoring continuously and allowing for a representation of both mild and chronic heart conditions. Despite this potential, relevant databases and benchmark studies are scarce. In this paper, we introduce our publicly accessible database, the Heart Sounds Shenzhen Corpus (HSS), which was first released during the recent INTERSPEECH 2018 ComParE Heart Sound sub-challenge. Additionally, we provide a survey of machine learning work in the area of heart sound recognition, as well as a benchmark for HSS utilising standard acoustic features and machine learning models. At best our support vector machine with Log Mel features achieves 49.7% unweighted average recall on a three category task (normal, mild, moderate/severe).