arXiv subjects

Xi Yin

Publications and source records attributed to Xi Yin.

At least 73 records · Page 4Linked to original sources

arXiv2019

Gait Recognition via Disentangled Representation Learning

Gait, the walking pattern of individuals, is one of the most important biometrics modalities. Most of the existing gait recognition methods take silhouettes or articulated body models as the gait features. These methods suffer from degraded recognition performance when handling confounding variables, such as clothing, carrying and view angle. To remedy this issue, we propose a novel AutoEncoder framework to explicitly disentangle pose and appearance features from RGB imagery and the LSTM-based integration of pose features over time produces the gait feature. In addition, we collect a Frontal-View Gait (FVG) dataset to focus on gait recognition from frontal-view walking, which is a challenging problem since it contains minimal gait cues compared to other views. FVG also includes other important variations, e.g., walking speed, carrying, and clothing. With extensive experiments on CASIA-B, USF and FVG datasets, our method demonstrates superior performance to the state of the arts quantitatively, the ability of feature disentanglement qualitatively, and promising computational efficiency.

cs.CV↗

arXiv2019

The c=1 String Theory S-Matrix Revisited

We revisit the perturbative S-matrix of c=1 string theory from the worldsheet perspective. We clarify the origin of the leg pole factors, the non-analyticity of the string amplitudes, and the validity as well as limitations of earlier computations based on resonance momenta. We compute the tree level 4-point amplitude and the genus one 2-point reflection amplitude by numerically integrating Virasoro conformal blocks with DOZZ structure constants on the sphere and on the torus, with sufficiently generic complex Liouville momenta, and find agreement with known answers from the c=1 matrix model.

hep-th↗

arXiv2019

Topological Defect Lines and Renormalization Group Flows in Two Dimensions

We consider topological defect lines (TDLs) in two-dimensional conformal field theories. Generalizing and encompassing both global symmetries and Verlinde lines, TDLs together with their attached defect operators provide models of fusion categories without braiding. We study the crossing relations of TDLs, discuss their relation to the 't Hooft anomaly, and use them to constrain renormalization group flows to either conformal critical points or topological quantum field theories (TQFTs). We show that if certain non-invertible TDLs are preserved along a RG flow, then the vacuum cannot be a non-degenerate gapped state. For various massive flows, we determine the infrared TQFTs completely from the consideration of TDLs together with modular invariance.

hep-th↗

arXiv2018

Genus Two Modular Bootstrap

We study the Virasoro conformal block decomposition of the genus two partition function of a two-dimensional CFT by expanding around a Z3-invariant Riemann surface that is a three-fold cover of the Riemann sphere branched at four points, and explore constraints from genus two modular invariance and unitarity. In particular, we find 'critical surfaces' that constrain the structure constants of a CFT beyond what is accessible via the crossing equation on the sphere.

hep-th↗

arXiv2018

Recursive Representations of Arbitrary Virasoro Conformal Blocks

We derive recursive representations in the internal weights of N-point Virasoro conformal blocks in the sphere linear channel and the torus necklace channel, and recursive representations in the central charge of arbitrary Virasoro conformal blocks on the sphere, the torus, and higher genus Riemann surfaces in the plumbing frame.

hep-th↗

arXiv2018

Long String Scattering in c = 1 String Theory

We study the scattering of long strings in c = 1 string theory, both in the worldsheet description and in the non-singlet sector of the dual matrix quantum mechanics. From the worldsheet perspective, the scattering amplitudes of long strings are obtained from a decoupling limit of open strings amplitudes on FZZT branes, which we compute by integrating Virasoro conformal blocks along with structure constants of boundary Liouville theory. In particular, we study the tree level amplitudes of (1) a long string decaying by emitting a closed string, and (2) the scattering of a pair of long strings. We show that they are indeed well defined as limits of open string amplitudes, and that our results are in striking numerical agreement with computations in the adjoint and bi-adjoint sectors of the dual matrix model (based on proposals of Maldacena and solutions due to Fidkowski), thereby providing strong evidence of the duality.

hep-th↗

arXiv2018

Representation Learning by Rotating Your Faces

The large pose discrepancy between two face images is one of the fundamental challenges in automatic face recognition. Conventional approaches to pose-invariant face recognition either perform face frontalization on, or learn a pose-invariant representation from, a non-frontal face image. We argue that it is more desirable to perform both tasks jointly to allow them to leverage each other. To this end, this paper proposes a Disentangled Representation learning-Generative Adversarial Network (DR-GAN) with three distinct novelties. First, the encoder-decoder structure of the generator enables DR-GAN to learn a representation that is both generative and discriminative, which can be used for face image synthesis and pose-invariant face recognition. Second, this representation is explicitly disentangled from other face variations such as pose, through the pose code provided to the decoder and pose estimation in the discriminator. Third, DR-GAN can take one or multiple images as the input, and generate one unified identity representation along with an arbitrary number of synthetic face images. Extensive quantitative and qualitative evaluation on a number of controlled and in-the-wild databases demonstrate the superiority of DR-GAN over the state of the art in both learning representations and rotating large-pose face images.

cs.CV↗

arXiv2017

Towards Large-Pose Face Frontalization in the Wild

Despite recent advances in face recognition using deep learning, severe accuracy drops are observed for large pose variations in unconstrained environments. Learning pose-invariant features is one solution, but needs expensively labeled large-scale data and carefully designed feature learning algorithms. In this work, we focus on frontalizing faces in the wild under various head poses, including extreme profile views. We propose a novel deep 3D Morphable Model (3DMM) conditioned Face Frontalization Generative Adversarial Network (GAN), termed as FF-GAN, to generate neutral head pose face images. Our framework differs from both traditional GANs and 3DMM based modeling. Incorporating 3DMM into the GAN structure provides shape and appearance priors for fast convergence with less training data, while also supporting end-to-end training. The 3DMM-conditioned GAN employs not only the discriminator and generator loss but also a new masked symmetry loss to retain visual quality under occlusions, besides an identity loss to recover high frequency information. Experiments on face recognition, landmark localization and 3D reconstruction consistently show the advantage of our frontalization method on faces in the wild datasets.

cs.CV↗

arXiv2017

Illuminating Pedestrians via Simultaneous Detection & Segmentation

Pedestrian detection is a critical problem in computer vision with significant impact on safety in urban autonomous driving. In this work, we explore how semantic segmentation can be used to boost pedestrian detection accuracy while having little to no impact on network efficiency. We propose a segmentation infusion network to enable joint supervision on semantic segmentation and pedestrian detection. When placed properly, the additional supervision helps guide features in shared layers to become more sophisticated and helpful for the downstream pedestrian detector. Using this approach, we find weakly annotated boxes to be sufficient for considerable performance gains. We provide an in-depth analysis to demonstrate how shared layers are shaped by the segmentation supervision. In doing so, we show that the resulting feature maps become more semantically meaningful and robust to shape and occlusion. Overall, our simultaneous detection and segmentation framework achieves a considerable gain over the state-of-the-art on the Caltech pedestrian dataset, competitive performance on KITTI, and executes 2x faster than competitive methods.

cs.CV↗

arXiv2017

Joint Multi-Leaf Segmentation, Alignment and Tracking from Fluorescence Plant Videos

This paper proposes a novel framework for fluorescence plant video processing. The plant research community is interested in the leaf-level photosynthetic analysis within a plant. A prerequisite for such analysis is to segment all leaves, estimate their structures, and track them over time. We identify this as a joint multi-leaf segmentation, alignment, and tracking problem. First, leaf segmentation and alignment are applied on the last frame of a plant video to find a number of well-aligned leaf candidates. Second, leaf tracking is applied on the remaining frames with leaf candidate transformation from the previous frame. We form two optimization problems with shared terms in their objective functions for leaf alignment and tracking respectively. A quantitative evaluation framework is formulated to evaluate the performance of our algorithm with four metrics. Two models are learned to predict the alignment accuracy and detect tracking failure respectively in order to provide guidance for subsequent plant biology analysis. The limitation of our algorithm is also studied. Experimental results show the effectiveness, efficiency, and robustness of the proposed method.

cs.CV↗

arXiv2017

Multi-Task Convolutional Neural Network for Pose-Invariant Face Recognition

This paper explores multi-task learning (MTL) for face recognition. We answer the questions of how and why MTL can improve the face recognition performance. First, we propose a multi-task Convolutional Neural Network (CNN) for face recognition where identity classification is the main task and pose, illumination, and expression estimations are the side tasks. Second, we develop a dynamic-weighting scheme to automatically assign the loss weight to each side task, which is a crucial problem in MTL. Third, we propose a pose-directed multi-task CNN by grouping different poses to learn pose-specific identity features, simultaneously across all poses. Last but not least, we propose an energy-based weight analysis method to explore how CNN-based MTL works. We observe that the side tasks serve as regularizations to disentangle the variations from the learnt identity features. Extensive experiments on the entire Multi-PIE dataset demonstrate the effectiveness of the proposed approach. To the best of our knowledge, this is the first work using all data in Multi-PIE for face recognition. Our approach is also applicable to in-the-wild datasets for pose-invariant face recognition and achieves comparable or better performance than state of the art on LFW, CFP, and IJB-A datasets.

cs.CV↗

arXiv2017

Bootstrapping the Spectral Function: On the Uniqueness of Liouville and the Universality of BTZ

We introduce spectral functions that capture the distribution of OPE coefficients and density of states in two-dimensional conformal field theories, and show that nontrivial upper and lower bounds on the spectral function can be obtained from semidefinite programming. We find substantial numerical evidence indicating that OPEs involving only scalar Virasoro primaries in a c>1 CFT are necessarily governed by the structure constants of Liouville theory. Combining this with analytic results in modular bootstrap, we conjecture that Liouville theory is the unique unitary c>1 CFT whose primaries have bounded spins. We also use the spectral function method to study modular constraints on CFT spectra, and discuss some implications of our results on CFTs of large c and large gap, in particular, to what extent the BTZ spectral density is universal.

hep-th↗

arXiv2016

(2,2) Superconformal Bootstrap in Two Dimensions

We find a simple relation between two-dimensional BPS N=2 superconformal blocks and bosonic Virasoro conformal blocks, which allows us to analyze the crossing equations for BPS 4-point functions in unitary (2,2) superconformal theories numerically with semidefinite programming. We constrain gaps in the non-BPS spectrum through the operator product expansion of BPS operators, in ways that depend on the moduli of exactly marginal deformations through chiral ring coefficients. In some cases, our bounds on the spectral gaps are observed to be saturated by free theories, by N=2 Liouville theory, and by certain Landau-Ginzburg models.

hep-th↗

arXiv2016

Modular Bootstrap Revisited

We constrain the spectrum of two-dimensional unitary, compact conformal field theories with central charge c > 1 using modular bootstrap. Upper bounds on the gap in the dimension of primary operators of any spin, as well as in the dimension of scalar primaries, are computed numerically as functions of the central charge using semi-definite programming. Our bounds refine those of Hellerman and Friedan-Keller, and are in some cases saturated by known CFTs. In particular, we show that unitary CFTs with c < 8 must admit relevant deformations, and that a nontrivial bound on the gap of scalar primaries exists for c < 25. We also study bounds on the dimension gap in the presence of twist gaps, bounds on the degeneracy of operators, and demonstrate how "extremal spectra" which maximize the degeneracy at the gap can be determined numerically.

hep-th↗

arXiv2015

N=4 Superconformal Bootstrap of the K3 CFT

We study two-dimensional (4,4) superconformal field theories of central charge c=6, corresponding to nonlinear sigma models on K3 surfaces, using the superconformal bootstrap. This is made possible through a surprising relation between the BPS N=4 superconformal blocks with c=6 and bosonic Virasoro conformal blocks with c=28, and an exact result on the moduli dependence of a certain integrated BPS 4-point function. Nontrivial bounds on the non-BPS spectrum in the K3 CFT are obtained as functions of the CFT moduli, that interpolate between the free orbifold points and singular CFT points. We observe directly from the CFT perspective the signature of a continuous spectrum above a gap at the singular moduli, and find numerically an upper bound on this gap that is saturated by the $A_1$ N=4 cigar CFT. We also derive an analytic upper bound on the first nonzero eigenvalue of the scalar Laplacian on K3 in the large volume regime, that depends on the K3 moduli data. As two byproducts, we find an exact equivalence between a class of BPS N=2 superconformal blocks and Virasoro conformal blocks in two dimensions, and an upper bound on the four-point functions of operators of sufficiently low scaling dimension in three and four dimensional CFTs.

hep-th↗

arXiv2015

Interpolating the Coulomb Phase of Little String Theory

We study up to 8-derivative terms in the Coulomb branch effective action of (1,1) little string theory, by collecting results of 4-gluon scattering amplitudes from both perturbative 6D super-Yang-Mills theory up to 4-loop order, and tree-level double scaled little string theory (DSLST). In previous work we have matched the 6-derivative term from the 6D gauge theory to DSLST, indicating that this term is protected on the entire Coulomb branch. The 8-derivative term, on the other hand, is unprotected. In this paper we compute the 8-derivative term by interpolating from the two limits, near the origin and near the infinity on the Coulomb branch, numerically from SU(k) SYM and DSLST respectively, for k=2,3,4,5. We discuss the implication of this result on the UV completion of 6D SYM as well as the strong coupling completion of DSLST. We also comment on analogous interpolating functions in the Coulomb phase of circle-compactified (2,0) little string theory.

hep-th↗

arXiv2015

On the Ground State Wave Function of Matrix Theory

We propose an explicit construction of the leading terms in the asymptotic expansion of the ground state wave function of BFSS SU(N) matrix quantum mechanics. Our proposal is consistent with the expected factorization property in various limits of the Coulomb branch, and involves a different scaling behavior from previous suggestions. We comment on some possible physical implications.

hep-th↗

arXiv2015

Supersymmetry Constraints and String Theory on K3

We study supervertices in six dimensional (2,0) supergravity theories, and derive supersymmetry non-renormalization conditions on the 4- and 6-derivative four-point couplings of tensor multiplets. As an application, we obtain exact non-perturbative results of such effective couplings in type IIB string theory compactified on K3 surface, extending previous work on type II/heterotic duality. The weak coupling limit thereof, in particular, gives certain integrated four-point functions of half-BPS operators in the nonlinear sigma model on K3 surface, that depend nontrivially on the moduli, and capture worldsheet instanton contributions.

hep-th↗