Search

Although interactive image segmentation techniques have made significant progress, supervised learning-based methods rely heavily on large-scale labeled data which is difficult to obtain in certain domains such as medicine, biology, etc. Models trained on natural images also struggle to achieve satisfactory results when directly applied to these domains. To solve this dilemma, we propose a Self-supervised Interactive Segmentation (SIS) method that achieves superior generalization performance. By clustering features from unlabeled data, we obtain classifiers that assign pseudo-labels to pixels in images. After refinement by super-pixel voting, these pseudo-labels are then used to train our segmentation network. To enable our network to better adapt to cross-domain images, we introduce correction learning and anti-forgetting regularization to conduct test-time adaptation. Our experiment results on five datasets show that our approach significantly outperforms other interactive segmentation methods across natural image datasets in the same conditions and achieves even better performance than some supervised methods when across to medical image domain. The code and models are available at https://github.com/leal0110/SIS.

ABSTRACT Biaxial and triaxial compression tests were performed on 100 x 100 x 100 mm cubic specimens of concrete under different stress loading rates. All the tests were performed using [...]

• A new method was proposed for non-stationary solutions to high-dimensional FPK equation. • Non-stationary stochastic responses of typical higher-dimensional dynamical systems have been analyzed. • Comparisons with the MCS method demonstrate the efficiency and accuracy of the proposed procedure. An approximate scheme is developed for efficiently determining non-stationary response PDF of multi-degree-of-freedom (MDOF) dynamical system to time-modulated random excitation. It further generalizes state-space-split exponential polynomial closure (SSS-EPC) technique and enhances its application for determining non-stationary response probability density function (PDF). To this aim, time variable is introduced into the approximate solution by setting polynomial coefficients as time dependent ones. With the proposed procedure, non-stationary solution to high-dimensional FPK equation can be obtained. Meanwhile, Monte Carlo simulation (MCS) method is conducted to assess the reliability of the proposed method. Three typical examples of linear and nonlinear MDOF dynamical systems to different modulated Gaussian white noise excitations are considered. Comparisons with the simulated results and computational time of the MCS method demonstrate the relatively high efficiency and accuracy of the proposed procedure. [ABSTRACT FROM AUTHOR]

• Comparisons between solutions and computational time evidence that the developed EPC method is an effective and efficient method. • The influence of correlation between excitations on system responses depends directly on the correlation coefficient. • The influence of correlation between excitations on system response seems stationary, and not affected by the time. • When the external and parametric excitations are correlated, the EQL method is not applicable since it does not reflect properties of system responses with unsymmetrical PDFs and nonzero means, especially for velocity response. In this paper, transient influence of correlation between external and parametric excitations on system response is investigated. Time variable has been taken into account. This does not seem to have been studied previously. By describing correlation between excitations quantitively with correlated coefficient, general Fokker-Planck-Kolmogorov (FPK) equation is reformulated with two parts. The first part relates to the one of independent excitations, while the other part is caused by the correlated excitations. Then, exponential polynomial closure (EPC) method, by taking time variable into account, is further developed for transient responses of nonlinear oscillators under correlated excitations. With the improved EPC method, typical system of Duffing oscillator under correlated external and parametric Gaussian white noise excitations is investigated. Based on the results, it is found the influence of correlation between excitations on system response depends directly on the magnitude and the sign of the correlation coefficient. Besides, the influence seems stationary, and not affected by the time. In addition, the results obtained from the EQL method are not consistent with the actual ones with unsymmetrical PDFs and nonzero means. It is indicated that the EQL method is not applicable to such case. [ABSTRACT FROM AUTHOR]

Statistical properties of random response essentially depend on the system and excitation features. The mathematical expressions for the response statistical properties should include the information on the system and excitation. In order to adequately illustrate the dependence of random response on the system and excitation features, only the stationary response is concerned. The exponential-polynomial-closure (EPC) method is further generalized and explicitly includes this information in the response probability density function (PDF) approximation. The unknown coefficients in the EPC approximate solution, which are constants in the previous stationary expression, are generalized to be functions of the system and excitation parameters. With the consideration of the independence among these parameters, each unknown coefficient is finally expressed as the product of several functions of each parameter. Based on the simulated response PDF results, the function of each parameter was determined by the least-squares method. Such a generalization deeply depicts the essentials of the response statistical properties and uniquely determines the response PDF expression. Three typical nonlinear systems are taken as examples to verify the efficiency of the proposed method. Numerical results show that approximate results obtained by the proposed method agree well with the simulated or exact ones. In addition, the evolution of the response PDF to the system or excitation parameters is obtained, which can be utilized for stochastic sensitivity analysis. • The dependence of the stationary response on the system and excitation features is investigated. • The unknown coefficients in the EPC approximation are generalized to be functions of system and excitation parameters. • The generalized coefficients are determined by the least square method based on the simulated results. • Numerical results show that results obtained by the proposed method agree well with the simulation or the exact solutions. [ABSTRACT FROM AUTHOR]