Domain adaptation proves to be an effective solution for addressing inadequate translation performance within specific domains. However, the straightforward approach of mixing data from multiple domains to obtain the multi-domain neural machine translation (NMT) model can give rise to the parameter interference between domains problem, resulting in a degradation of overall performance. To address this, we introduce a multi-domain adaptive NMT method aimed at learning domain specific sub-layer latent variable and employ the Gumbel-Softmax reparameterization technique to concurrently train both model parameters and domain specific sub-layer latent variable. This approach facilitates learning private domain-specific knowledge while sharing common domain-invariant knowledge, effectively mitigating the parameter interference problem. The experimental results show that our proposed method significantly improved by up to 7.68 and 3.71 BLEU compared with the baseline model in English-German and Chinese-English public multi-domain datasets, respectively.

Generating high-quality stitched images with natural structures is a challenging task in computer vision. In this paper, we succeed in preserving both local and global geometric structures for wide parallax images, while reducing artifacts and distortions. A projective invariant, Characteristic Number, is used to match co-planar local sub-regions for input images. The homography between these well-matched sub-regions produces consistent line and point pairs, suppressing artifacts in overlapping areas. We explore and introduce global collinear structures into an objective function to specify and balance the desired characters for image warping, which can preserve both local and global structures while alleviating distortions. We also develop comprehensive measures for stitching quality to quantify the collinearity of points and the discrepancy of matched line pairs by considering the sensitivity to linear structures for human vision. Extensive experiments demonstrate the superior performance of the proposed method over the state-of-the-art by presenting sharp textures and preserving prominent natural structures in stitched images. Especially, our method not only exhibits lower errors but also the least divergence across all test images. Code is available at https://github.com/dut-media-lab/Image-Stitching.

Animation generating, which refers to generating a movie or animation according to one or more images, has a number of useful application ranging from movie effects to virtual reality. This paper proposes a novel animation generating method via MRLS image deformation and designs an animation system, which can generate a vividly animation from a static image by choosing set of point handles and deformation paths. Firstly, users need to choose a set of fixed control points and a distorted image deformation path. Then, construct deformation mapping functions according to the control points and the deformation path, and get a series of deformed images with the mapping function. Finally, users get the animation by packaging the deformed images into an AVI video files frame-by-frame. Experiment results show that the non-rigid deformation animations produced with our method are deformed smoothly, naturally and realistically.

Dynamic constrained multi-objective differential evolution (DCMODE) is designed for solving constrained optimization problem (COP). Main feature presented in this paper is to construct dynamic multi-objective optimization problem (DMOP) from COP. The two evolved objectives are original function objective and violation objective. Constraints are controlled by dynamic environments, where the relaxed constraints boundaries are gradually tightened to original boundaries. After this dynamic process, DMOP solutions are close to COP solution. This new algorithm is tested on benchmark problems of special session at CEC2006 with 100% success rates of all problems. Compared with several state-of-the-art DE variants referred in this paper, our algorithm outperforms or performs similarly to them. The satisfactory results suggest that it is efficient and generic when handling inequality/equality constraints.

Constrained optimization problem (COP) is skillfully converted into dynamic constrained multi-objective optimization problem (DCMOP) in this paper. Then dynamic constrained multi-objective evolutionary algorithms (DCMOEAs) can be used to solve the COP problem by solving the DCMOP problem. Seemingly, a complex DCMOEA algorithm is used to solve a relatively simple COP problem. However, the DCMOEA algorithm can adopt Pareto domination to achieve a good trade off between fast converging and global searching, and therefore a DCMOEA algorithm can effectively solve a COP problem by solving the DCMOP problem. An instance of DCMOEA was used to to solve 13 widely used constraint benchmark problems, The experimental results suggest it outperforms or performs similarly to other state-of-the-art algorithms referred to in this paper. The efficient performance of the DCMOEA algorithm shows, to some extend, the DCMOP model works well.