Stacked generative adversarial networks for image compositing

Perfect image compositing can harmonize the appearance between the foreground and background effectively so that the composite result looks seamless and natural. However, the traditional convolutional neural network (CNN)-based methods often fail to yield highly realistic composite results due to overdependence on scene parsing while ignoring the coherence of semantic and structural between foreground and background. In this paper, we propose a framework to solve this problem by training a stacked generative adversarial network with attention guidance, which can efficiently create a high-resolution, realistic-looking composite. To this end, we develop a diverse adversarial loss in addition to perceptual and guidance loss to train the proposed generative network. Moreover, we construct a multi-scenario dataset for high-resolution image compositing, which contains high-quality images with different styles and object masks. Experiments on the synthesized and real images demonstrate the efficiency and effectiveness of our network in producing seamless, natural, and realistic results. Ablation studies show that our proposed network can improve the visual performance of composite results compared with the application of existing methods.

mSwap: a large-scale image-compositing method with optimal m-ary tree

With the increasing of computing ability, large-scale simulations have been generating massive amounts of data in aerodynamics. Sort-last parallel rendering is the most classical image compositing method for large-scale scientific visualization. However, in the stage of image compositing, the sort-last method may suffer from scalability problem on large-scale processors. Existing image compositing algorithms tend to perform well in certain situations. For instance, Direct Send is well on small and medium scale; Radix-k gets well performance only when the k-value is appropriate and so on. In this paper, we propose a novel method named mSwap for scientific visualization in aerodynamics, which uses the best scale of processors to make sure its performance at the best. mSwap groups the processors that we can use with a (m,k) table, which records the best combination of m (the number of processors in subgroup of each group) and k (the number of processors in each group). Then in each group, using a m-ary tree to composite the image for reducing the communication of processors. Finally, the image is composited between different groups to generate the final image. The performance and scalability of our mSwap method is demonstrated through experiments with thousands of processors.

Modified Poisson compositing technique on skewed grid

<abstract><p>Image compositing is the process of seamlessly inserting a portion of a source image into a target image to create a new desirable image. This work describes an image composition approach based on numerical differentiation utilizing the Laplacian operator. The suggested procedure uses the red-black strategy to speed up computations by using two separate relaxation factors for red and black nodes, as well as two accelerated parameters on a skewed grid. The Skewed Modified Two-Parameter Overrelaxation (SkMTOR) approach is a modification of the existing MTOR method. The SkMTOR has been used to solve numerous linear equations in the past, but its applicability in image processing has never been investigated. Several examples were used to test the suggested method in solving the Poisson equation for image composition. The results demonstrated that the image composition was successfully constructed using all six methods considered in this study. The six methods evaluated yielded identical images based on the similarity measurement results. In terms of computing speed, the skewed variants perform much quicker than their corresponding regular grid variants, with the SkMTOR showing the best performance.</p></abstract>

mSwap: A Large-Scale Image-Compositing Method with Optimal m-ary Tree

With the increasing of computing ability, large-scale simulations have been generating massive amounts of data in aerodynamics. Sort-last parallel rendering is the most classical image compositing method for large-scale scientific visualization. However, in the stage of image compositing, the sort-last method may suffer from scalability problem on large-scale processors. Existing image compositing algorithms tend to perform well in certain situations. For instance, Direct Send is well on small and medium scale; Radix-k gets well performance only when the k-value is appropriate and so on. In this paper, we propose a novel method named mSwap for scientific visualization in aerodynamics, which uses the best scale of processors to make sure its performance at the best. mSwap groups the processors that we can use with a (m, k) table, which records the best combination of m (the number of processors in subgroup of each group) and k (the number of processors in each group). Then in each group, using a m-ary tree to composite the image for reducing the communication of processors. Finally, the image is composited between different groups to generate the final image. The performance and scalability of our mSwap method is demonstrated through experiments with thousands of processors.

mSwap: A Large-Scale Image-Compositing Method with Optimal m-ary Tree

With the increasing of computing ability, large-scale science simulations have been generating massive amounts of data in aerodynamics. Sort-last parallel rendering is a proven approach for large-scale science visualization. However, in the stage of image compositing, the sort-last method may suffer from scalability problem on large-scale processors. Existing image compositing algorithms tend to perform well in certain situations. For instance, Direct Send is well on small and medium scale; Radix-k gets well performance only when the k -value is appropriate and so on. In this paper, we propose a novel method named mSwap for science visualization in aerodynamics, which uses the best scale of processors to make sure its performance at the best. mSwap groups the processors that we can use with a ( m, k ) table, which records the best combination of m (the number of processors in subgroup of each group) and k (the number of processors in each group). Then in each group, using a m-ary tree to composite the image for reducing the communication of processors. Finally, the image is composited between different groups to generate the final image. The performance and scalability of our mSwap method is demonstrated through experiments with thousands of processors.