In this work, we present a domain flow generation(DLOW) model to bridge two different domains by generating a continuous sequence of intermediate domains flowing from one domain to the other. The benefits of our DLOW model are two-fold. First, it is able to transfer source images into different styles in the intermediate domains. The transferred images smoothly bridge the gap between source and target domains, thus easing the domain adaptation task. Second, when multiple target domains are provided for training, our DLOW model is also able to generate new styles of images that are unseen in the training data. We implement our DLOW model based on CycleGAN. A domainness variable is introduced to guide the model to generate the desired intermediate domain images. In the inference phase, a flow of various styles of images can be obtained by varying the domainness variable. We demonstrate the effectiveness of our model for both cross-domain semantic segmentation and the style generalization tasks on benchmark datasets. Our implementation is available at

Perfectly hydrophobic (PHO) coatings consisting of silicone nanofibers have been obtained via a solution process using methyltrialkoxysilanes as precursors. On the basis of thermal gravimetry and differential thermal analysis (TG-DTA) and Fourier transform infrared spectroscopy (FTIR) results, the formula of the nanofibers was tentatively given and a possible growth mechanism of the nanofibers was proposed. Because of the low affinity between the coatings and the small water droplet, when using these coatings as substrate for collecting water vapor, the harvesting efficiency could be enhanced as compared with those from bare glass substrate for more than 50% under 25 °C and 60-90% relative humidity. By removing the surface methyl group by heat treatment or ultraviolet (UV) irradiation, the as-prepared perfectly hydrophobic surface can be converted into a superhydrophilic surface.

Open compound domain adaptation (OCDA) is a domain adaptation setting, where target domain is modeled as a compound of multiple unknown homogeneous domains, which brings the advantage of improved generalization to unseen domains. In this work, we propose a principled meta-learning based approach to OCDA for semantic segmentation, MOCDA, by modeling the unlabeled target domain continuously. Our approach consists of four key steps. First, we cluster target domain into multiple sub-target domains by image styles, extracted in an unsupervised manner. Then, different sub-target domains are split into independent branches, for which batch normalization parameters are learnt to treat them independently. A meta-learner is thereafter deployed to learn to fuse sub-target domainspecific predictions, conditioned upon the style code. Meanwhile, we learn to online update the model by modelagnostic meta-learning (MAML) algorithm, thus to further improve generalization. We validate the benefits of our approach by extensive experiments on synthetic-to-real knowledge transfer benchmark datasets, where we achieve the state-of-the-art performance in both compound and open domains.

No abstract

The three-dimensional, inviscid and viscous flow instability modes that appear on a solid-body rotation flow in a finite-length straight, circular pipe are analysed. This study is a direct extension of the Wang & Rusak (Phys. Fluids, vol. 8 (4), 1996a, pp. 1007–1016) analysis of axisymmetric instabilities on inviscid swirling flows in a pipe. The linear stability equations are the same as those derived by Kelvin (Phil. Mag., vol. 10, 1880, pp. 155–168). However, we study a general mode of perturbation that satisfies the inlet, outlet and wall conditions of a flow in a finite-length pipe with a fixed in time and in space vortex generator ahead of it. This mode is different from the classical normal mode of perturbations. The eigenvalue problem for the growth rate and the shape of the perturbations for any azimuthal wavenumber $m$ consists of a linear system of partial differential equations in terms of the axial and radial coordinates ($x,r$). The stability problem is solved numerically for all azimuthal wavenumbers $m$. The computed growth rates and the related shapes of the various perturbation modes that appear in sequence as a function of the base flow swirl ratio (${\it\omega}$) and pipe length ($L$) are presented. In the inviscid flow case, the $m=1$ modes are the first to become unstable as the swirl ratio is increased and dominate the perturbation’s growth in a certain range of swirl levels. The $m=1$ instability modes compete with the axisymmetric ($m=0$) instability modes as the swirl ratio is further increased. In the viscous flow case, the viscous damping effects reduce the modes’ growth rates. The neutral stability line is presented in a Reynolds number ($Re$) versus swirl ratio (${\it\omega}$) diagram and can be used to predict the first appearance of axisymmetric or spiral instabilities as a function of $Re$ and $L$. We use the Reynolds–Orr equation to analyse the various production terms of the perturbation’s kinetic energy and establish the elimination of the flow axial homogeneity at high swirl levels as the underlying physical mechanism that leads to flow exchange of stability and to the appearance of both spiral and axisymmetric instabilities. The viscous effects in the bulk have only a passive influence on the modes’ shapes and growth rates. These effects decrease with the increase of $Re$. We show that the inviscid flow stability results are the inviscid-limit stability results of high-$Re$ rotating flows.

This work presents a robust, and low-cost framework for real-time marker based 3-D human expression modeling using off-the-shelf stereo web-cameras and inexpensive adhesive markers applied to the face. The system has low computational requirements, runs on standard hardware, and is portable with minimal setup time and no training. It does not require a controlled lab environment (lighting or setup) and is robust under varying conditions, i.e. illumination, facial hair, or skin tone variation. Stereo web-cameras perform 3-D marker tracking to obtain head rigid motion and the non-rigid motion of expressions. Tracked markers are then mapped onto a 3-D face model with a virtual muscle animation system. Muscle inverse kinematics update muscle contraction parameters based on marker motion in order to create a virtual character's expression performance. The parametrization of the muscle-based animation encodes a face performance with little bandwidth. Additionally, a radial basis function mapping approach was used to easily remap motion capture data to any face model. In this way the automated creation of a personalized 3-D face model and animation system from 3-D data is elucidated. The expressive power of the system and its ability to recognize new expressions was evaluated on a group of test subjects with respect to the six universally recognized facial expressions. Results show that the use of abstract muscle definition reduces the effect of potential noise in the motion capture data and allows the seamless animation of any virtual anthropomorphic face model with data acquired through human face performance.

The physical properties of a recently proposed feedback-stabilization method of a vortex flow in a finite-length straight pipe are studied for the case of a solid-body rotation flow. In the natural case, when the swirl ratio is beyond a certain critical level, linearly unstable modes appear in sequence as the swirl level is increased. Based on an asymptotic long-wave (long-pipe) approach, the global feedback control method is shown to enforce the decay in time of the perturbation’s kinetic energy and thereby quench all of the instability modes for a swirl range above the critical swirl level. The effectiveness of an extended version of this feedback flow control approach is further analysed through a detailed mode analysis of the full linear control problem for a solid-body rotation flow in a finite-length pipe that is not necessarily long. We first rigourously prove the asymptotic decay in time of all modes with real growth rates. We then compute the growth rate and shape of all modes according to the full linearized control problem for swirl levels up to 50 % above the critical level. We demonstrate that the flow is stabilized in the whole swirl range and can be even further stabilized for higher swirl levels. However, the control effectiveness is sensitive to the choice of the feedback control gain. A potentially best range of the gain is identified. An inadequate level of gain, either insufficient or excessive, could lead to a marginal control or failure of the control method at high swirl levels. The robustness of the proposed control law to stabilize both initial waves and continuous inlet flow perturbations and the elimination of the vortex breakdown process are demonstrated through numerical computations.

One challenge of object recognition is to generalize to new domains, to more classes and/or to new modalities. This necessitates methods to combine and reuse existing datasets that may belong to different domains, have partial annotations, and/or have different data modalities. This paper formulates this as a multi-source domain adaptation and label unification problem, and proposes a novel method for it. Our method consists of a partially-supervised adaptation stage and a fully-supervised adaptation stage. In the former, partial knowledge is transferred from multiple source domains to the target domain and fused therein. Negative transfer between unmatching label spaces is mitigated via three new modules: domain attention, uncertainty maximization and attention-guided adversarial alignment. In the latter, knowledge is transferred in the unified label space after a label completion process with pseudo-labels. Extensive experiments on three different tasks -image classification, 2D semantic image segmentation, and joint 2D-3D semantic segmentation -show that our method outperforms all competing methods significantly.