arXiv subjects

Bin Feng

Publications and source records attributed to Bin Feng.

69 records · Page 4Linked to original sources

arXiv2021

Optofluidic ptychography on a chip

We report the implementation of a fully on-chip, lensless microscopy technique termed optofluidic ptychography. This imaging modality complements the miniaturization provided by microfluidics and allows the integration of ptychographic microscopy into various lab-on-a-chip devices. In our prototype, we place a microfluidic channel on the top surface of a coverslip and coat the bottom surface with a scattering layer. The channel and the coated coverslip substrate are then placed on top of an image sensor for diffraction data acquisition. Similar to the operation of flow cytometer, the device utilizes microfluidic flow to deliver specimens across the channel. The diffracted light from the flowing objects is modulated by the scattering layer and recorded by the image sensor for ptychographic reconstruction, where high-resolution quantitative complex images are recovered from the diffraction measurements. By using an image sensor with a 1.85-micron pixel size, our device can resolve the 550 nm linewidth on the resolution target. We validate the device by imaging different types of biospecimens, including C. elegans, yeast cells, paramecium, and closterium sp. We also demonstrate high-resolution ptychographic reconstruction at a video framerate of 30 frames per second. The reported technique can address a wide range of biomedical needs and engenders new ptychographic imaging innovations in a flow cytometer configuration.

physics.ins-det↗

arXiv2021

Tracking Instances as Queries

Recently, query based deep networks catch lots of attention owing to their end-to-end pipeline and competitive results on several fundamental computer vision tasks, such as object detection, semantic segmentation, and instance segmentation. However, how to establish a query based video instance segmentation (VIS) framework with elegant architecture and strong performance remains to be settled. In this paper, we present \textbf{QueryTrack} (i.e., tracking instances as queries), a unified query based VIS framework fully leveraging the intrinsic one-to-one correspondence between instances and queries in QueryInst. The proposed method obtains 52.7 / 52.3 AP on YouTube-VIS-2019 / 2021 datasets, which wins the 2-nd place in the YouTube-VIS Challenge at CVPR 2021 \textbf{with a single online end-to-end model, single scale testing \& modest amount of training data}. We also provide QueryTrack-ResNet-50 baseline results on YouTube-VIS-2021 val set as references for the VIS community.

cs.CV↗

arXiv2021

Instances as Queries

Recently, query based object detection frameworks achieve comparable performance with previous state-of-the-art object detectors. However, how to fully leverage such frameworks to perform instance segmentation remains an open problem. In this paper, we present QueryInst (Instances as Queries), a query based instance segmentation method driven by parallel supervision on dynamic mask heads. The key insight of QueryInst is to leverage the intrinsic one-to-one correspondence in object queries across different stages, as well as one-to-one correspondence between mask RoI features and object queries in the same stage. This approach eliminates the explicit multi-stage mask head connection and the proposal distribution inconsistency issues inherent in non-query based multi-stage instance segmentation methods. We conduct extensive experiments on three challenging benchmarks, i.e., COCO, CityScapes, and YouTube-VIS to evaluate the effectiveness of QueryInst in instance segmentation and video instance segmentation (VIS) task. Specifically, using ResNet-101-FPN backbone, QueryInst obtains 48.1 box AP and 42.8 mask AP on COCO test-dev, which is 2 points higher than HTC in terms of both box AP and mask AP, while runs 2.4 times faster. For video instance segmentation, QueryInst achieves the best performance among all online VIS approaches and strikes a decent speed-accuracy trade-off. Code is available at \url{https://github.com/hustvl/QueryInst}.

cs.CV↗

arXiv2021

Crossover Learning for Fast Online Video Instance Segmentation

Modeling temporal visual context across frames is critical for video instance segmentation (VIS) and other video understanding tasks. In this paper, we propose a fast online VIS model named CrossVIS. For temporal information modeling in VIS, we present a novel crossover learning scheme that uses the instance feature in the current frame to pixel-wisely localize the same instance in other frames. Different from previous schemes, crossover learning does not require any additional network parameters for feature enhancement. By integrating with the instance segmentation loss, crossover learning enables efficient cross-frame instance-to-pixel relation learning and brings cost-free improvement during inference. Besides, a global balanced instance embedding branch is proposed for more accurate and more stable online instance association. We conduct extensive experiments on three challenging VIS benchmarks, \ie, YouTube-VIS-2019, OVIS, and YouTube-VIS-2021 to evaluate our methods. To our knowledge, CrossVIS achieves state-of-the-art performance among all online VIS methods and shows a decent trade-off between latency and accuracy. Code will be available to facilitate future research.

cs.CV↗

arXiv2020

Deep multi-metric learning for text-independent speaker verification

Text-independent speaker verification is an important artificial intelligence problem that has a wide spectrum of applications, such as criminal investigation, payment certification, and interest-based customer services. The purpose of text-independent speaker verification is to determine whether two given uncontrolled utterances originate from the same speaker or not. Extracting speech features for each speaker using deep neural networks is a promising direction to explore and a straightforward solution is to train the discriminative feature extraction network by using a metric learning loss function. However, a single loss function often has certain limitations. Thus, we use deep multi-metric learning to address the problem and introduce three different losses for this problem, i.e., triplet loss, n-pair loss and angular loss. The three loss functions work in a cooperative way to train a feature extraction network equipped with Residual connections and squeeze-and-excitation attention. We conduct experiments on the large-scale \texttt{VoxCeleb2} dataset, which contains over a million utterances from over $6,000$ speakers, and the proposed deep neural network obtains an equal error rate of $3.48\%$, which is a very competitive result. Codes for both training and testing and pretrained models are available at \url{https://github.com/GreatJiweix/DmmlTiSV}, which is the first publicly available code repository for large-scale text-independent speaker verification with performance on par with the state-of-the-art systems.

eess.AS↗

arXiv2020

Maximum Entropy Regularization and Chinese Text Recognition

Chinese text recognition is more challenging than Latin text due to the large amount of fine-grained Chinese characters and the great imbalance over classes, which causes a serious overfitting problem. We propose to apply Maximum Entropy Regularization to regularize the training process, which is to simply add a negative entropy term to the canonical cross-entropy loss without any additional parameters and modification of a model. We theoretically give the convergence probability distribution and analyze how the regularization influence the learning process. Experiments on Chinese character recognition, Chinese text line recognition and fine-grained image classification achieve consistent improvement, proving that the regularization is beneficial to generalization and robustness of a recognition model.

cs.CV↗

arXiv2019

Patch Aggregator for Scene Text Script Identification

Script identification in the wild is of great importance in a multi-lingual robust-reading system. The scripts deriving from the same language family share a large set of characters, which makes script identification a fine-grained classification problem. Most existing methods make efforts to learn a single representation that combines the local features by making a weighted average or other clustering methods, which may reduce the discriminatory power of some important parts in each script for the interference of redundant features. In this paper, we present a novel module named Patch Aggregator (PA), which learns a more discriminative representation for script identification by taking into account the prediction scores of local patches. Specifically, we design a CNN-based method consisting of a standard CNN classifier and a PA module. Experiments demonstrate that the proposed PA module brings significant performance improvements over the baseline CNN model, achieving the state-of-the-art results on three benchmark datasets for script identification: SIW-13, CVSI 2015 and RRC-MLT 2017.

cs.CV↗

arXiv2019

High electrical conducting deep-ultraviolet-transparent oxide semiconductor La-doped SrSnO3 exceeding ~3000 S cm-1

La-doped SrSnO3 (LSSO) is known as one of deep-ultraviolet (DUV)-transparent conducting oxides with an energy bandgap of ~4.6 eV. Since LSSO can be grown heteroepitaxially on more wide bandgap substrates such as MgO (Eg ~7.8 eV), LSSO is considered to be a good candidate as a DUV-transparent electrode. However, the electrical conductivity of LSSO films are below 1000 S cm^-1, most likely due to the low solubility of La ion in the LSSO lattice. Here we report that high electrically conducting (>3000 S cm^-1) LSSO thin films with an energy bandgap of ~4.6 eV can be fabricated by pulsed laser deposition on MgO substrate followed by a simple annealing in vacuum. From the X-ray diffraction and the scanning transmission electron microscopy analyses, we found that lateral grain growth occurred during the annealing, which improved the activation rate of La ion, leading to a significant improvement of carrier concentration (3.26 x 10^20 cm^-3) and Hall mobility (55.8 cm^2 V^-1 s^-1). The present DUV-transparent oxide semiconductor would be useful as a transparent electrode for developing optoelectronic devices, which transmit and/or emit DUV-light.

cond-mat.mtrl-sci↗

arXiv2018

Buffer layer-less fabrication of high-mobility transparent oxide semiconductor, La-doped BaSnO3

Transparent oxide semiconductors (TOSs) showing both high visible transparency and high electron mobility have attracted great attention towards the realization of advanced optoelectronic devices. La-doped BaSnO3 (LBSO) is one of the most promising TOSs because its single crystal exhibits a high electron mobility. However, in the LBSO films, it is very hard to obtain high mobility due to the threading dislocations, which are originated from the lattice mismatch between the film and the substrate. Therefore, many researchers have tried to improve the mobility by inserting a buffer layer. While the buffer layers increased the electron mobilities, this approach leaves much to be desired since it involves a two-step film fabrication process and the enhanced mobility values are still significantly lower than single crystal values. We show herein that the electron mobility of LBSO films can be improved without inserting any buffer layers if the films are grown under highly oxidative ozone (O3) atmospheres. The O3 environments relaxed the LBSO lattice and reduced the formation of Sn2+ states, which are known to suppress the electron mobility in LBSO. The resultant O3-LBSO films showed improved mobility values up to 115 cm2 V-1 s-1, which is among the highest in LBSO films on SrTiO3 substrates and comparable to LBSO films with buffer layers.

cond-mat.mtrl-sci↗

arXiv2018

Large thickness dependence of the carrier mobility in a transparent oxide semiconductor, La-doped BaSnO3

We report herein that the carrier mobility of the 2%-La-doped BaSnO3 (LBSO) films on (001) SrTiO3 and (001) MgO substrates strongly depends on the thickness whereas it is unrelated to the lattice mismatch (+5.4% for SrTiO3, -2.3% for MgO). Although we observed large differences in the lattice parameters, the lateral grain size (~85 nm for SrTiO3, ~20 nm for MgO), the surface morphology and the density of misfit dislocations, the mobility increased almost simultaneously with the thickness in both cases and saturated at ~100 cm2 V-1 s-1, together with the approaching to the nominal carrier concentration (=[2% La3+]), clearly indicating that the behavior of mobility depends on the film thickness. The present results would be beneficial to understand the behavior of mobility and fruitful to further enhance the mobility of LBSO films.

cond-mat.mtrl-sci↗

arXiv2017

Atomic-Scale Structure Relaxation, Chemistry and Charge Distribution of Dislocation Cores in SrTiO3

By using the state-of-the-art microscopy and spectroscopy in aberration-corrected scanning transmission electron microscopes, we determine the atomic arrangements, occupancy, elemental distribution, and the electronic structures of dislocation cores in the 10°tilted SrTiO3 bicrystal. We identify that there are two different types of oxygen deficient dislocation cores, i.e., the SrO plane terminated Sr0.82Ti0.85O3-x (Ti3.67+, 0.48<x<0.91) and TiO2 plane terminated Sr0.63Ti0.90O3-y (Ti3.60+, 0.57<y<1). They have the same Burgers vector of a[100] but different atomic arrangements and chemical properties. Besides the oxygen vacancies, Sr vacancies and rocksalt-like titanium oxide reconstruction are also identified in the dislocation core with TiO2 plane termination. Our atomic-scale study reveals the true atomic structures and chemistry of individual dislocation cores, providing useful insights into understanding the properties of dislocations and grain boundaries.

cond-mat.mtrl-sci↗

arXiv2017

Thermoelectric phase diagram of the SrTiO3-SrNbO3 solid solution system

Thermoelectric energy conversion - the exploitation of the Seebeck effect to convert waste heat into electricity - has attracted an increasing amount of research attention for energy harvesting technology. Niobium-doped strontium titanate (SrTi1-xNbxO3) is one of the most promising thermoelectric material candidates, particularly as it poses a much lesser environmental risk in comparison to materials based on heavy metal elements. Two-dimensional electron confinement, e.g. through the formation of superlattices or two-dimensional electron gases, is recognized as an effective strategy to improve the thermoelectric performance of SrTi1-xNbxO3. Although electron confinement is closely related to the electronic structure, the fundamental electronic phase behavior of the SrTi1-xNbxO3 solid solution system has yet to be comprehensively investigated. Here, we present a thermoelectric phase diagram for the SrTi1-xNbxO3 (0.05 =< x =< 1) solid solution system, which we derived from the characterization of epitaxial films. We observed two thermoelectric phase boundaries in the system, which originate from the step-like decrease in carrier effective mass at x ~ 0.3, and from a local minimum in carrier relaxation time at x ~ 0.5. The origins of these phase boundaries are considered to be related to isovalent/heterovalent B-site substitution: parabolic Ti 3d orbitals dominate electron conduction for compositions with x < 0.3, whereas the Nb 4d orbital dominates when x > 0.3. At x ~ 0.5, a tetragonal distortion of the lattice, in which the B-site is composed of Ti4+ and Nb4+ ions, leads to the formation of tail-like impurity bands, which maximizes the electron scattering. These results provide a foundation for further research into improving the thermoelectric performance of SrTi1-xNbxO3.

cond-mat.mtrl-sci↗

arXiv2017

Thermopower modulation clarification of the intrinsic effective mass in a transparent oxide semiconductor, BaSnO3

Although there are so many reports on the carrier effective mass (m*) of a transparent oxide semiconductor BaSnO3, it is almost impossible to know the intrinsic m* value because the reported m* values are scattered from 0.06 to 3.7 m0. Here we successfully clarified the intrinsic m* of BaSnO3, m*=0.40 0.01 m0, by the thermopower modulation clarification method. We also found the threshold of degenerate/non-degenerate semiconductor of BaSnO3; At the threshold, the thermopower value of both La-doped BaSnO3 and BaSnO3 TFT structure was 240 microvolt k-1, bulk carrier concentration was 1.4E19 cm-3, and two-dimensional sheet carrier concentration was 1.8E12 cm-2. When the EF locates above the parabolic shaped conduction band bottom, rather high mobility was observed. On the contrary, very low carrier mobility was observed when the EF lays below the threshold, most likely due to that the tail states suppress the carrier mobility. The present results are useful for further development of BaSnO3 based oxide electronics.

cond-mat.mtrl-sci↗

arXiv2011

Suppression of FM-to-AM conversion in third-harmonic generation by tuning the ratio of modulation depth

Issues of Frequency-to-Amplitude modulation (FM-to-AM) conversion occurred in phase-modulated third-harmonic generation (THG) process are investigated. An expression about group-velocity is theoretically derived to suppress the FM-to-AM conversion, which appears to be dependant on the ratio of modulation depth of fundamental to second-harmonic when given the same modulation frequencies of them. Simulation results indicate that the induced AM in THG process can be suppressed effectively when the expression about group-velocity is satisfied.

physics.optics↗

arXiv2011

Suppression of FM-to-AM conversion in third-harmonic generation at the retracing point of a crystal

FM-to-AM conversion can cause many negative effects (e.g., reducing of margin against damage to the optics, and etc.) on performances of third-harmonic conversion system. In this letter, the FM-to-AM conversion effect in third-harmonic generation is investigated both at and away from the retracing point of type-II KDP crystal. Obtained results indicate that the FM-to-AM conversion can be suppressed effectively when the crystal works at the retracing point.

physics.optics↗