Search

Recent years have seen many advances based on Deep Convolutional Neural Networks (DCNNs) in the tasks of face recognition, most of which are developed to pursue high recognition accuracy. In this paper, we propose a novel Fast FAce Recognizer (Fast-FAR), learning to improve the speed of DCNN-based face recognition model without sacrificing recognition accuracy. Our fundamental insight is that the computation increases exponentially with the depth of a network, the easily identifiable face images can be accurately recognized by the cheap features (pixel values at shallow layers), while the challenging samples that exhibit low quality, large pose variations or occlusions need to be processed by the expensive deep layers. The major contribution of this paper is the Reinforcement Learning Agent (RLA), which is proposed to learn a decision policy determined by a reward function. The policy adaptively decides whether the recognition should be performed at an early layer with a high recognition confidence, or proceeding to the subsequent layers, thus significantly reducing feed-forward cost for the easy faces. According to the extensive experiments on the popular face recognition benchmarks, Fast-FAR reduces the inference time by 14.22%, 20.61%, and 7.84% on the dataset LFW, AgeDB-30 and CFP-FP, respectively.

Reduced graphene oxide has attracted numerous interests due to its unique, superior electronic, optical, mechanical, and chemical properties. An epoxy resin with excellent mechanical and electrical properties can be obtained by doping with reduced graphene oxide to enhance the function of the polymer. Here, we prepared a uniform reduced graphene oxide/epoxy resin coating with a different reduced graphene oxide content and characterized it using a field-emission scanning electron microscope (FE-SEM), X-ray diffractometer (XRD), Raman, and Fourier transform infrared spectrometer (FTIR). Furthermore, the spectral characteristics of the composite coating in the terahertz band were discussed. The cross-sectional SEM results show that a fold structure with ductile failure was intensively formed due to the compatibility of graphene and polymer materials. Both the Raman G and Raman 2D peaks of reduced graphene oxide were confirmed using Raman spectrum testing. The diffraction peak of reduced graphene oxide at 24° disappeared within the reduced graphene oxide/epoxy resin coating, and a wide diffraction peak of the amorphous structure was formed together. Additionally, the intensity of the Raman spectrum increased significantly with increased reduced graphene oxide content, thereby making the surface electrical resistance of the coatings decrease exponentially. Additionally, the intensity of the terahertz time-domain signal and frequency-domain power spectrum linearly reduced with increased reduced graphene oxide concentration. However, the terahertz absorption coefficient and refractive index both increased gradually with increased reduced graphene oxide doping due to increased orientation polarization in the composite coating.

We present a versatile model, FaceAnime, for various video generation tasks from still images. Video generation from a single face image is an interesting problem and usually tackled by utilizing Generative Adversarial Networks (GANs) to integrate information from the input face image and a sequence of sparse facial landmarks. However, the generated face images usually suffer from quality loss, image distortion, identity change, and expression mismatching due to the weak representation capacity of the facial landmarks. In this paper, we propose to "imagine" a face video from a single face image according to the reconstructed 3D face dynamics, aiming to generate a realistic and identity-preserving face video, with precisely predicted pose and facial expression. The 3D dynamics reveal changes of the facial expression and motion, and can serve as a strong prior knowledge for guiding highly realistic face video generation. In particular, we explore face video prediction and exploit a well-designed 3D dynamic prediction network to predict a 3D dynamic sequence for a single face image. The 3D dynamics are then further rendered by the sparse texture mapping algorithm to recover structural details and sparse textures for generating face frames. Our model is versatile for various AR/VR and entertainment applications, such as face video retargeting and face video prediction. Superior experimental results have well demonstrated its effectiveness in generating high-fidelity, identity-preserving, and visually pleasant face video clips from a single source face image.

In real-world scenarios, many factors may harm face recognition performance, e.g., large pose, bad illumination,low resolution, blur and noise. To address these challenges, previous efforts usually first restore the low-quality faces to high-quality ones and then perform face recognition. However, most of these methods are stage-wise, which is sub-optimal and deviates from the reality. In this paper, we address all these challenges jointly for unconstrained face recognition. We propose an Multi-Degradation Face Restoration (MDFR) model to restore frontalized high-quality faces from the given low-quality ones under arbitrary facial poses, with three distinct novelties. First, MDFR is a well-designed encoder-decoder architecture which extracts feature representation from an input face image with arbitrary low-quality factors and restores it to a high-quality counterpart. Second, MDFR introduces a pose residual learning strategy along with a 3D-based Pose Normalization Module (PNM), which can perceive the pose gap between the input initial pose and its real-frontal pose to guide the face frontalization. Finally, MDFR can generate frontalized high-quality face images by a single unified network, showing a strong capability of preserving face identity. Qualitative and quantitative experiments on both controlled and in-the-wild benchmarks demonstrate the superiority of MDFR over state-of-the-art methods on both face frontalization and face restoration., 14 pages, 9 figures

3D face reconstruction from a single 2D image is a very important topic in computer vision. However, the current reconstruction methods are usually non-sensitive to face identities and over-sensitive to facial poses, which may result in similar 3D geometries for faces of different identities, or obtain different shapes for the same identity with different poses. When such methods are applied practically, their 3D estimates are either changeable for different photos of the same subject or over-regularized and generic to distinguish face identities. In this paper, we propose a robust solution to solve this problem by carefully designing a novel Siamese Convolutional Neural Network (SCNN). Specifically, regarding the 3D Morphable face Model (3DMM) parameters of the same individual as the same class, we employ the contrastive loss to enlarge the inter-class distance and meanwhile reduce the intra-class distance for the output 3DMM parameters. We also propose an identity loss to preserve the identity information for the same individual in the feature space. Training with these two losses, our SCNN could learn representations that are more discriminative for face identity and generalizable for pose variants. Experiments on the challenging database 300W-LP and AFLW2000-3D have shown the effectiveness of our method by comparing with state-of-the-arts., 5 pages, 6 figures, IEEE International Conference on Information Communication and Signal Processing

Silicon photonics integrated circuits (Si-PIC) with well-established active and passive building elements are progressing towards large-scale commercialization in optical communications and high speed optical interconnects applications. However, current Si-PICs do not have memory capabilities, in particular, the non-volatile memory functionality for energy efficient data storage. Here, we propose an electrically programmable, multi-level non-volatile photonics memory cell (PMC) fabricated by standard complementary-metal-oxide-semiconductor (CMOS) compatible processes. A micro-ring resonator (MRR) was built using the PMC to optically read the memory states. Switching energy smaller than 20 pJ was achieved. Additionally, a MRR memory array was employed to demonstrate a four-bit memory read capacity. Theoretically, this can be increased up to ~400 times using a 100 nm free spectral range broadband light source. The fundamental concept of this design provides a route to eliminate the von Neumann bottleneck. The energy-efficient optical storage can complement on-chip optical interconnects for neutral networking, memory input/output interfaces and other computational intensive applications.

The first pn-type carrier-induced silicon Bragg-grating filter is demonstrated. The extinction-ratio modulations are 11.5 dB and 10 dB with reverse and forward biases, respectively. 8-Gpbs data rate is achieved with a reverse bias., {"references":["H. Tsuda, \"Fiber Bragg grating vibration-sensing system, insensitive to Bragg wavelength and employing fiber ring laser,\" Opt. Lett. 35, pp. 2349-2351 (2010).","Y. G. Han and S. B. Lee, \"Tunable dispersion compensator based on uniform fiber Bragg grating and its application to tunable pulse repetition-rate multiplication,\" Opt. Express 13, pp. 9224-9229 (2005).","D. T. H. Tan, et al., \"Cladding-modulated Bragg grating in silicon waveguides,\" Opt. Lett. 34, pp. 1357-1359 (2006).","S. Honda, et al., \"Largely-tunable wideband Bragg gratings fabricated on SOI rib waveguides employed by deep-RIE,\" Electron. Lett. 43, pp. 630-631 (2007).","G. Jiang, et al., \"Slab-modulated sidewall Bragg grating in silicon-on-insulator ridge waveguides,\" IEEE Photon. Technol. 23, pp. 6-8 (2011).","Q. Fang, et al., \"Carrier-induced silicon Bragg grating filters with a p-i-n junction,\" IEEE Photon. Technol. 25, pp. 810-812 (2013)."]}

Despite all the eminent advantages of silicon photonics, other materials need to be integrated to fulfill the functions that are difficult to realize with silicon alone. This is because silicon has a low light emission efficiency and a low electro-optic coefficient, limiting the use of silicon as a material for light sources and modulators. A strong two-photon absorption (TPA) at high intensities also limits the use of silicon in applications exploiting nonlinear effects. In addition, signal amplification is needed to compensate the insertion and propagation losses in silicon nanowaveguides. To address these issues we have demonstrated the integration of atomic layer deposited nanolaminates on silicon waveguides. Firstly we demonstrate slot waveguide ring resonators patterned on a silicon-on-insulator (SOI) wafer coated with an atomic layer deposited organic/inorganic nanolaminate structure, which consists of alternating layers of tantalum pentoxide (Ta2O5) and polyimide (PI) [1]. These materials were selected since the ALD process for depositing Ta2O5/PI nanolaminate films is already available [2] and both materials exhibit high third order nonlinearities [3-4]. In our nanolaminate ring resonators, the optical power is not only confined in the narrow central air slot but also in several parallel sub-10 nm wide vertical polyimide slots. This indicates that the mode profiles in the silicon slot waveguide can be accurately tuned by the atomic layer deposition (ALD) method. Our results show that ALD of organic and inorganic materials can be combined with conventional silicon waveguide fabrication techniques to create slot waveguide ring resonators with varying mode profiles. Secondly we demonstrate the integration of atomic layer deposited erbium-doped aluminum oxide (Al2O3) nanolaminates on silicon waveguides. This method provides an efficient way for controlling the concentration and distribution of erbium ions. We have applied this method on silicon strip and slot waveguides and show signal enhancement. Our results show that atomic layer deposited nanolaminates can potentially open new possibilities for various photonic applications, such as silicon photonic devices for light emission and amplification, optical sensing and all-optical signal processing. References [1] A. Autere, L. Karvonen, A. Saynatjoki, M. Roussey, E. Farm, M. Kemell, X. Tu, T.Y. Liow, G.Q. Lo, M. Ritala, M. Leskela, S. Honkanen, H. Lipsanen, and Z. Sun, "Slot waveguide ring resonators coated by an atomic layer deposited organic/inorganic nanolaminate," Opt. Express 23, 26940-26951 (2015) [2] L. D. Salmi, E. Puukilainen, M. Vehkamaki, M. Heikkila, and M. Ritala, “Atomic layer deposition of Ta2O5/polyimide nanolaminates,” Chem. Vap. Deposition 15, 221–226 (2009). [3] S. Morino, T. Yamashita, K. Horie, T. Wada, and H. Sasabe, “Third-order nonlinear optical properties of aromatic polyisoimides,” React. Funct. Polym. 44, 183–188 (2000). [4] C.-Y. Tai, J. Wilkinson, N. Perney, M. Netti, F. Cattaneo, C. Finlayson, and J. Baumberg, “Determination of nonlinear refractive index in a Ta2O5 rib waveguide using self-phase modulation,” Opt. Express 12, 5110–5116 (2004).

We demonstrate a 1×4 coarse wavelength division-multiplexing (CWDM) planar concave grating multiplexer/demultiplexer and its application in re-configurable optical add/drop multiplexer (ROADM) system in silicon-on-insulator substrate. The wavelengths of the demonstrated concave grating multiplexer align well with the ITU-T standard. We demonstrate a prototype of ROADM comprising two such concave gratings and four wide-band thermo-optical MZI switches. Undercut technology which removes the underneath silicon substrate is adopted in optical switches in order to minimize the operation power. For all the thermal heaters, the operation voltage is smaller than 1.5 V, and the switch power is ~2.4 mW. High throughput pseudorandom binary sequence (PRBS) data transmission with up to 100 Gb/s is demonstrated, showing the high-performance ROADM functionality.

Different types of slot waveguide couplers are studied theoretically and experimentally. We present strip-to-slot waveguide couplers with a length as small as 10 μm and with their feature sizes no less than 150 nm, exhibiting efficient coupling into the slot mode and negligible scattering. We demonstrate coupling loss of 0.11 dB for a 20-micron long coupler from a strip waveguide to slot waveguide, fabricated using the 248 nm deep-UV lithography. We also discuss ring resonator couplers, multimode multislot structures for 2x2 couplers, and prism coupling characterization of multislot waveguide structures.

We demonstrate silicon-based multiple-slot waveguides filled with dual atomic layer deposited oxide layers. Slot modes for both polarizations, transverse electric (TE) and transverse magnetic (TM), are supported in the waveguide. Propagation loss in the order of 8 dB/cm is achieved for the TE-polarization and 4 dB/cm for the TM-polarization in the waveguides with dual (Al 2 O 3 -TiO 2 ) thin film layers. The devices are fabricated using low-temperature complementary metal-oxide-semiconductor compatible processes. To our knowledge, this is the first demonstration of dual-filled slot waveguides.

We demonstrate low-loss silicon slot waveguides filled with single and dual atomic layer deposited oxide layers. Propagation losses less than 5 dB/cm and 8 dB/cm are achieved for the waveguides with single (Al 2 O 3 ) and double (Al 2 O 3 -TiO 2 ) layers, respectively. The devices are fabricated using low-temperature CMOS compatible processes. The geometries allow nonlinearities nearly two orders of magnitude smaller than plain silicon waveguides.