No abstract

No abstract

Referring image segmentation aims to segment the objects referred by a natural language expression. Previous methods usually focus on designing an implicit and recurrent feature interaction mechanism to fuse the visuallinguistic features to directly generate the final segmentation mask without explicitly modeling the localization information of the referent instances. To tackle these problems, we view this task from another perspective by decoupling it into a "Locate-Then-Segment" (LTS) scheme. Given a language expression, people generally first perform attention to the corresponding target image regions, then generate a fine segmentation mask about the object based on its context. The LTS first extracts and fuses both visual and textual features to get a cross-modal representation, then applies a cross-model interaction on the visual-textual features to locate the referred object with position prior, and finally generates the segmentation result with a light-weight segmentation network. Our LTS is simple but surprisingly effective. On three popular benchmark datasets, the LTS outperforms all the previous state-of-the-arts methods by a large margin (e.g., +3.2% on RefCOCO+ and +3.4% on RefCOCOg). In addition, our model is more interpretable with explicitly locating the object, which is also proved by visualization experiments. We believe this framework is promising to serve as a strong baseline for referring image segmentation.

No abstract

A novel back-propagation algorithm is proposed based on perturbation analysis to compensate the intra-channel nonlinearity. It reduces 85% back-propagation stages in an experiment of dual polarization QPSK transmission over 1500 km dispersion uncompensated link.

No abstract

The phase noise induced by the intra-channel nonlinearity with DP-16QAM is found to be higher than that with DP-QPSK in 1600km transmission experiments. The main reason is identified as the large IXPM in 16QAM based on the proposed nonlinear model. IntroductionWithout inline dispersion compensation, nondispersion managed (NDM) system appears to have better nonlinear tolerance both for interchannel and intra-channel nonlinearities

Linear scan computed tomography (CT) is a promising imaging configuration with high scanning efficiency while the data set is under-sampled and angularly limited for which high quality image reconstruction is challenging. In this work, an edge guided total variation minimization reconstruction (EGTVM) algorithm is developed in dealing with this problem. The proposed method is modeled on the combination of total variation (TV) regularization and iterative edge detection strategy. In the proposed method, the edge weights of intermediate reconstructions are incorporated into the TV objective function. The optimization is efficiently solved by applying alternating direction method of multipliers. A prudential and conservative edge detection strategy proposed in this paper can obtain the true edges while restricting the errors within an acceptable degree. Based on the comparison on both simulation studies and real CT data set reconstructions, EGTVM provides comparable or even better quality compared to the non-edge guided reconstruction and adaptive steepest descent-projection onto convex sets method. With the utilization of weighted alternating direction TV minimization and edge detection, EGTVM achieves fast and robust convergence and reconstructs high quality image when applied in linear scan CT with under-sampled data set.