Searcharxiv⌕ Search

arXiv subjects

Cheng Zhao

Publications and source records attributed to Cheng Zhao.

At least 127 records · Page 7Linked to original sources

SPP-Net: Deep Absolute Pose Regression with Synthetic Views

Image based localization is one of the important problems in computer vision due to its wide applicability in robotics, augmented reality, and autonomous systems. There is a rich set of methods described in the literature how to geometrically register a 2D image w.r.t.\ a 3D model. Recently, methods based on deep (and convolutional) feedforward networks (CNNs) became popular for pose regression. However, these CNN-based methods are still less accurate than geometry based methods despite being fast and memory efficient. In this work we design a deep neural network architecture based on sparse feature descriptors to estimate the absolute pose of an image. Our choice of using sparse feature descriptors has two major advantages: first, our network is significantly smaller than the CNNs proposed in the literature for this task---thereby making our approach more efficient and scalable. Second---and more importantly---, usage of sparse features allows to augment the training data with synthetic viewpoints, which leads to substantial improvements in the generalization performance to unseen poses. Thus, our proposed method aims to combine the best of the two worlds---feature-based localization and CNN-based pose regression--to achieve state-of-the-art performance in the absolute pose estimation. A detailed analysis of the proposed architecture and a rigorous evaluation on the existing datasets are provided to support our method.

cs.CV↗

Baryon Acoustic Oscillation detections from the clustering of massive halos and different density region tracers in TianNu simulation

The Baryon Acoustic Oscillations (BAO) refer to the ripples of material density in the Universe. As the most direct density tracers in the universe, galaxies have been commonly used in studies of BAO peak detection. The spatial number density of galaxies, to a certain extent, reflects the distribution of the material density of our Universe. Using galaxies as matter tracers, we can construct more overlapping empty spheres (DT voids) than the matter tracers, via Delaunay Triangulation technique. We show that their radii excellently reflect the galaxy number density round them, and they can serve as reliable different density region tracers. Using the data from an unprecedented large-scale $N$-body simulation "TianNu", we conduct some fundamental statistical studies and clustering analysis of the DT voids. We discuss in detail the representative features of two-point correlation functions of different DT void populations. We show that the peak, the position of which corresponds to the average radius of data samples, is the most representative feature of the two-point correlation function of the DT voids. In addition, we also construct another voids, the disjoint voids, and investigate their some statistical properties and clustering properties. And we find that the occupied space of all disjoint voids accounts for about $45\%$ of the volume of the simulation box, regardless of the number density of mock galaxies. We also investigate the BAO detections based on different tracers, i.e. mock galaxies, low-density region tracers, and high-density region tracers respectively. Our results show that BAO intensities detected by low/high-density region tracers are enhanced significantly compared to the BAO detection by mock galaxies, for the mock galaxy catalogue with the number density of $7.52\times10^{-5}$ $h^3$ Mpc$^{-3}$.

astro-ph.CO↗

Effect of Massive Neutrinos on the Position of Cold Dark Matter Halo: Revealed via Delaunay Triangulation Void

Using cosmological $N$-body simulation which coevolves cold dark matter (CDM) and neutrino particles, we discover the local effect of massive neutrinos on the spatial distribution of CDM halos, reflected on properties of the Delaunay Triangulation (DT) voids. Smaller voids are generally in regions with higher neutrino abundance and so their surrounding halos are impacted by a stronger neutrino free streaming. This makes the voids larger (surrounding halos being washed outward the void center). On the contrary, larger voids are generally in regions with lower neutrino abundance and so their surrounding halos are less impacted by neutrino free streaming, making the voids smaller (surrounding halos being squeezed toward the void center). This characteristic change of the spatial distribution of the halos suppresses the 2-point correlation function of halos on scales $\sim$ 1 Mpc$/h$ and significantly skews the number function of the DT voids, which serve as measurable neutrino effects in current or future galaxy surveys.

astro-ph.CO↗

The clustering of the SDSS-IV extended Baryon Oscillation Spectroscopic Survey DR14 quasar sample: First measurement of Baryon Acoustic Oscillations between redshift 0.8 and 2.2

We present measurements of the Baryon Acoustic Oscillation (BAO) scale in redshift-space using the clustering of quasars. We consider a sample of 147,000 quasars from the extended Baryon Oscillation Spectroscopic Survey (eBOSS) distributed over 2044 square degrees with redshifts $0.8 < z < 2.2$ and measure their spherically-averaged clustering in both configuration and Fourier space. Our observational dataset and the 1400 simulated realizations of the dataset allow us to detect a preference for BAO that is greater than 2.8$σ$. We determine the spherically averaged BAO distance to $z = 1.52$ to 3.8 per cent precision: $D_V(z=1.52)=3843\pm147 \left(r_{\rm d}/r_{\rm d, fid}\right)\ $Mpc. This is the first time the location of the BAO feature has been measured between redshifts 1 and 2. Our result is fully consistent with the prediction obtained by extrapolating the Planck flat $Λ$CDM best-fit cosmology. All of our results are consistent with basic large-scale structure (LSS) theory, confirming quasars to be a reliable tracer of LSS, and provide a starting point for numerous cosmological tests to be performed with eBOSS quasar samples. We combine our result with previous, independent, BAO distance measurements to construct an updated BAO distance-ladder. Using these BAO data alone and marginalizing over the length of the standard ruler, we find $Ω_Λ > 0$ at 6.6$σ$ significance when testing a $Λ$CDM model with free curvature.

astro-ph.CO↗

Dense RGB-D semantic mapping with Pixel-Voxel neural network

For intelligent robotics applications, extending 3D mapping to 3D semantic mapping enables robots to, not only localize themselves with respect to the scene's geometrical features but also simultaneously understand the higher level meaning of the scene contexts. Most previous methods focus on geometric 3D reconstruction and scene understanding independently notwithstanding the fact that joint estimation can boost the accuracy of the semantic mapping. In this paper, a dense RGB-D semantic mapping system with a Pixel-Voxel network is proposed, which can perform dense 3D mapping while simultaneously recognizing and semantically labelling each point in the 3D map. The proposed Pixel-Voxel network obtains global context information by using PixelNet to exploit the RGB image and meanwhile, preserves accurate local shape information by using VoxelNet to exploit the corresponding 3D point cloud. Unlike the existing architecture that fuses score maps from different models with equal weights, we proposed a Softmax weighted fusion stack that adaptively learns the varying contributions of PixelNet and VoxelNet, and fuses the score maps of the two models according to their respective confidence levels. The proposed Pixel-Voxel network achieves the state-of-the-art semantic segmentation performance on the SUN RGB-D benchmark dataset. The runtime of the proposed system can be boosted to 11-12Hz, enabling near to real-time performance using an i7 8-cores PC with Titan X GPU.

cs.CV↗

Accurate halo-galaxy mocks from automatic bias estimation and particle mesh gravity solvers

Reliable extraction of cosmological information from clustering measurements of galaxy surveys requires estimation of the error covariance matrices of observables. The accuracy of covariance matrices is limited by our ability to generate sufficiently large number of independent mock catalogs that can describe the physics of galaxy clustering across a wide range of scales. Furthermore, galaxy mock catalogs are required to study systematics in galaxy surveys and to test analysis tools. In this investigation, we present a fast and accurate approach for generation of mock catalogs for the upcoming galaxy surveys. Our method relies on low-resolution approximate gravity solvers to simulate the large scale dark matter field, which we then populate with halos according to a flexible nonlinear and stochastic bias model. In particular, we extend the \textsc{patchy} code with an efficient particle mesh algorithm to simulate the dark matter field (the \textsc{FastPM} code), and with a robust MCMC method relying on the \textsc{emcee} code for constraining the parameters of the bias model. Using the halos in the BigMultiDark high-resolution $N$-body simulation as a reference catalog, we demonstrate that our technique can model the bivariate probability distribution function (counts-in-cells), power spectrum, and bispectrum of halos in the reference catalog. Specifically, we show that the new ingredients permit us to reach percentage accuracy in the power spectrum up to $k\sim 0.4\; \,h\,{\rm Mpc}^{-1}$ (within 5\% up to $k\sim 0.6\; \,h\,{\rm Mpc}^{-1}$) with accurate bispectra improving previous results based on Lagrangian perturbation theory.

astro-ph.CO↗

EZmocks: extending the Zel'dovich approximation to generate mock galaxy catalogues with accurate clustering statistics

We present a new methodology to generate mock halo or galaxy catalogues, which have accurate clustering properties, nearly indistinguishable from full $N$-body solutions, in terms of the one-point, two-point, and three-point statistics. In particular, the agreement is remarkable, within $1\%$ up to $k=0.55$ $h$Mpc$^{-1}$ and down to $r=10$ $h^{-1}$Mpc, for the power spectrum and two-point correlation function respectively, while the bispectrum agrees in general within $20\%$ for different scales and shapes. Our approach is based on the Zel'dovich approximation, however, effectively including with the simple prescriptions the missing physical ingredients, and stochastic scale-dependent, non-local and nonlinear biasing contributions. The computing time and memory required to produce one mock is similar to that using the log-normal model. With high accuracy and efficiency, the effective Zel'dovich approximation mocks (EZmocks) provide a reliable and practical method to produce massive mock galaxy catalogues for the analysis of large-scale structure measurements.

astro-ph.CO↗

Linear redshift space distortions for cosmic voids based on galaxies in redshift space

Cosmic voids found in galaxy surveys are defined based on the galaxy distribution in redshift space. We show that the large scale distribution of voids in redshift space traces the fluctuations in the dark matter density field δ(k) (in Fourier space with μbeing the line of sight projected k-vector): δ_v^s(k) = (1 + β_v μ^2) b^s_v δ(k), with a beta factor that will be in general different than the one describing the distribution of galaxies. Only in case voids could be assumed to be quasi-local transformations of the linear (Gaussian) galaxy redshift space field, one gets equal beta factors β_v=β_g=f/b_g with f being the growth rate, and b_g, b^s_v being the galaxy and void bias on large scales defined in redshift space. Indeed, in our mock void catalogs we measure void beta factors being in good agreement with the galaxy one. Further work needs to be done to confirm the level of accuracy of the beta factor equality between voids and galaxies, but in general the void beta factor needs to be considered as a free parameter for RSD studies.

astro-ph.CO↗

Weakly-supervised DCNN for RGB-D Object Recognition in Real-World Applications Which Lack Large-scale Annotated Training Data

This paper addresses the problem of RGBD object recognition in real-world applications, where large amounts of annotated training data are typically unavailable. To overcome this problem, we propose a novel, weakly-supervised learning architecture (DCNN-GPC) which combines parametric models (a pair of Deep Convolutional Neural Networks (DCNN) for RGB and D modalities) with non-parametric models (Gaussian Process Classification). Our system is initially trained using a small amount of labeled data, and then automatically prop- agates labels to large-scale unlabeled data. We first run 3D- based objectness detection on RGBD videos to acquire many unlabeled object proposals, and then employ DCNN-GPC to label them. As a result, our multi-modal DCNN can be trained end-to-end using only a small amount of human annotation. Finally, our 3D-based objectness detection and multi-modal DCNN are integrated into a real-time detection and recognition pipeline. In our approach, bounding-box annotations are not required and boundary-aware detection is achieved. We also propose a novel way to pretrain a DCNN for the depth modality, by training on virtual depth images projected from CAD models. We pretrain our multi-modal DCNN on public 3D datasets, achieving performance comparable to state-of-the-art methods on Washington RGBS Dataset. We then finetune the network by further training on a small amount of annotated data from our novel dataset of industrial objects (nuclear waste simulants). Our weakly supervised approach has demonstrated to be highly effective in solving a novel RGBD object recognition application which lacks of human annotations.

cs.CV↗

A fully end-to-end deep learning approach for real-time simultaneous 3D reconstruction and material recognition

This paper addresses the problem of simultaneous 3D reconstruction and material recognition and segmentation. Enabling robots to recognise different materials (concrete, metal etc.) in a scene is important for many tasks, e.g. robotic interventions in nuclear decommissioning. Previous work on 3D semantic reconstruction has predominantly focused on recognition of everyday domestic objects (tables, chairs etc.), whereas previous work on material recognition has largely been confined to single 2D images without any 3D reconstruction. Meanwhile, most 3D semantic reconstruction methods rely on computationally expensive post-processing, using Fully-Connected Conditional Random Fields (CRFs), to achieve consistent segmentations. In contrast, we propose a deep learning method which performs 3D reconstruction while simultaneously recognising different types of materials and labelling them at the pixel level. Unlike previous methods, we propose a fully end-to-end approach, which does not require hand-crafted features or CRF post-processing. Instead, we use only learned features, and the CRF segmentation constraints are incorporated inside the fully end-to-end learned system. We present the results of experiments, in which we trained our system to perform real-time 3D semantic reconstruction for 23 different materials in a real-world application. The run-time performance of the system can be boosted to around 10Hz, using a conventional GPU, which is enough to achieve real-time semantic reconstruction using a 30fps RGB-D camera. To the best of our knowledge, this work is the first real-time end-to-end system for simultaneous 3D reconstruction and material recognition.

cs.CV↗

On the Capability of PID Control for Nonlinear Uncertain Systems

It is well-known that the classical PID controller is by far the most widely used ones in industrial processes, despite of the remarkable progresses of the modern control theory over the past half a century. It is also true that the existing theoretical studies on PID control mainly focus on linear systems, although most of the practical control systems are inherently nonlinear with uncertainties. Thus, a natural question is: can we establish a theory on PID controller for nonlinear uncertain dynamical systems? This paper will initiate an investigation on this fundamental problem, showing that any second order uncertain nonlinear dynamical systems can be stabilized globally by the PID controller as long as the nonlinearity satisfies a Lipschitz condition. We will also demonstrate that this result can be generalized neither to systems with order higher than 2, and nor to systems with nonlinear growth rate faster than linear in general.

math.OC↗

Signatures of the Primordial Universe from Its Emptiness: Measurement of Baryon Acoustic Oscillations from Minima of the Density Field

Sound waves from the primordial fluctuations of the Universe imprinted in the large-scale structure, called baryon acoustic oscillations (BAOs), can be used as standard rulers to measure the scale of the Universe. These oscillations have already been detected in the distribution of galaxies. Here we propose to measure BAOs from the troughs (minima) of the density field. Based on two sets of accurate mock halo catalogues with and without BAOs in the seed initial conditions, we demonstrate that the BAO signal cannot be obtained from the clustering of classical disjoint voids, but is clearly detected from overlapping voids. The latter represent an estimate of all troughs of the density field. We compute them from the empty circumsphere centers constrained by tetrahedra of galaxies using Delaunay triangulation. Our theoretical models based on an unprecedented large set of detailed simulated void catalogues are remarkably well confirmed by observational data. We use the largest recently publicly available sample of luminous red galaxies from SDSS-III BOSS DR11 to unveil for the first time a >3$σ$ BAO detection from voids in observations. Since voids are nearly isotropically expanding regions, their centers represent the most quiet places in the Universe, keeping in mind the cosmos origin and providing a new promising window in the analysis of the cosmological large-scale structure from galaxy surveys.

astro-ph.CO↗

The clustering of galaxies in the SDSS-III Baryon Oscillation Spectroscopic Survey: mock galaxy catalogues for the BOSS Final Data Release

We reproduce the galaxy clustering catalogue from the SDSS-III Baryon Oscillation Spectroscopic Survey Final Data Release (BOSS DR11&DR12) with high fidelity on all relevant scales in order to allow a robust analysis of baryon acoustic oscillations and redshift space distortions. We have generated (6,000) 12,288 MultiDark PATCHY BOSS (DR11) DR12 light-cones corresponding to an effective volume of $\sim192,000\,[h^{-1}\,{\rm Gpc}]^3$ (the largest ever simulated volume), including cosmic evolution in the redshift range from 0.15 to 0.75. The mocks have been calibrated using a reference galaxy catalogue based on the halo abundance matching modelling of the BOSS DR11&DR12 galaxy clustering data and on the data themselves. The production follows three steps. First, we apply the PATCHY code to generate a dark matter field and an object distribution including nonlinear stochastic galaxy bias. Secondly, we run the halo/stellar distribution reconstruction HADRON code to assign masses to the various objects. This step uses the mass distribution as a function of local density and non-local indicators (i.e., tidal field tensor eigenvalues and relative halo exclusion separation for massive objects) from the reference simulation applied to the corresponding PATCHY dark matter and galaxy distribution. Finally, we apply the SUGAR code to build the light cones. The resulting MultiDark PATCHY mock light cones reproduce the number density, selection function, survey geometry, and in general within 1 $σ$, for arbitrary stellar mass bins, the power spectrum up to $k=0.3\,h\,{\rm Mpc}^{-1}$, the two-point correlation functions down to a few Mpc scales, and the three-point statistics of the BOSS DR11&DR12 galaxy samples.

astro-ph.CO↗

DIVE in the cosmic web: voids with Delaunay Triangulation from discrete matter tracer distributions

We present a novel parameter-free cosmological void finder (\textsc{dive}, Delaunay TrIangulation Void findEr) based on Delaunay Triangulation (DT), which efficiently computes the empty spheres constrained by a discrete set of tracers. We define the spheres as DT voids, and describe their properties, including an universal density profile together with an intrinsic scatter. We apply this technique on 100 halo catalogues with volumes of 2.5\,$h^{-1}$Gpc side each, with a bias and number density similar to the BOSS CMASS Luminous Red Galaxies, performed with the \textsc{patchy} code. Our results show that there are two main species of DT voids, which can be characterised by the radius: they have different responses to halo redshift space distortions, to number density of tracers, and reside in different dark matter environments. Based on dynamical arguments using the tidal field tensor, we demonstrate that large DT voids are hosted in expanding regions, whereas the haloes used to construct them reside in collapsing ones. Our approach is therefore able to efficiently determine the troughs of the density field from galaxy surveys, and can be used to study their clustering. We further study the power spectra of DT voids, and find that the bias of the two populations are different, demonstrating that the small DT voids are essentially tracers of groups of haloes.

astro-ph.CO↗

Measuring Baryon Acoustic Oscillations from the clustering of voids

We investigate the necessary methodology to optimally measure the baryon acoustic oscillation (BAO) signal, from voids based on galaxy redshift catalogues. To this end, we study the dependency of the BAO signal on the population of voids classified by their sizes. We find for the first time the characteristic features of the correlation function of voids including the first robust detection of BAOs in mock galaxy catalogues. These show an anti-correlation around the scale corresponding to the smallest size of voids in the sample (the void exclusion effect), and dips at both sides of the BAO peak, which can be used to determine the significance of the BAO signal without any priori model. Furthermore, our analysis demonstrates that there is a scale dependent bias for different populations of voids depending on the radius, with the peculiar property that the void population with the largest BAO significance corresponds to tracers with approximately zero bias on the largest scales. We further investigate the methodology on an additional set of 1,000 realistic mock galaxy catalogues reproducing the SDSS-III/BOSS CMASS DR11 data, to control the impact of sky mask and radial selection function. Our solution is based on generating voids from randoms including the same survey geometry and completeness, and a post-processing cleaning procedure in the holes and at the boundaries of the survey. The methodology and optimal selection of void populations validated in this work have been used to perform the first BAO detection from voids in observations, presented in a companion paper.

astro-ph.CO↗

nIFTy Cosmology: Galaxy/halo mock catalogue comparison project on clustering statistics

We present a comparison of major methodologies of fast generating mock halo or galaxy catalogues. The comparison is done for two-point and the three-point clustering statistics. The reference catalogues are drawn from the BigMultiDark N-body simulation. Both friend-of-friends (including distinct halos only) and spherical overdensity (including distinct halos and subhalos) catalogs have been used with the typical number density of a large-volume galaxy surveys. We demonstrate that a proper biasing model is essential for reproducing the power spectrum at quasilinear and even smaller scales. With respect to various clustering statistics a methodology based on perturbation theory and a realistic biasing model leads to very good agreement with N-body simulations. However, for the quadrupole of the correlation function or the power spectrum, only the method based on semi-N-body simulation could reach high accuracy (1% level) at small scales, i.e., r<25 Mpc/h or k>0.15 h/Mpc. Full N-body solutions will remain indispensable to produce reference catalogues. Nevertheless, we have demonstrated that the far more efficient approximate solvers can reach a few percent accuracy in terms of clustering statistics at the scales interesting for the large-scale structure analysis after calibration with a few reference N-body calculations. This makes them useful for massive production aimed at covariance studies, to scan large parameter spaces, and to estimate uncertainties in data analysis techniques, such as baryon acoustic oscillation reconstruction, redshift distortion measurements, etc.

astro-ph.CO↗

Halo mass distribution reconstruction across the cosmic web

We study the relation between halo mass and its environment from a probabilistic perspective. We find that halo mass depends not only on local dark matter density, but also on non-local quantities such as the cosmic web environment and the halo-exclusion effect. Given these accurate relations, we have developed the HADRON-code (Halo mAss Distribution ReconstructiON), a technique which permits us to assign halo masses to a distribution of haloes in three-dimensional space. This can be applied to the fast production of mock galaxy catalogues, by assigning halo masses, and reproducing accurately the bias for different mass cuts. The resulting clustering of the halo populations agree well with that drawn from the BigMultiDark $N$-body simulation: the power spectra are within 1-$σ$ up to scales of $k=0.2\,h\,{\rm Mpc}^{-1}$, when using augmented Lagrangian perturbation theory based mock catalogues. Only the most massive haloes show a larger deviation. For these, we find evidence of the halo-exclusion effect. A clear improvement is achieved when assigning the highest masses to haloes with a minimum distance separation. We also compute the 2- and 3-point correlation functions, and find an excellent agreement with $N$-body results. Our work represents a quantitative application of the cosmic web classification. It can have further interesting applications in the multi-tracer analysis of the large-scale structure for future galaxy surveys.

astro-ph.CO↗

Hunting down systematics in baryon acoustic oscillations after cosmic high noon

Future dark energy experiments will require better and more accurate theoretical predictions for the baryonic acoustic oscillations (BAO) signature in the spectrum of cosmological perturbations. Here, we use large N-body simulations of the \LambdaCDM Planck cosmology to study any possible systematic shifts and damping in BAO due to the impact of nonlinear gravitational growth of structure, scale dependent and non-local bias, and redshift-space distortions. The effect of cosmic variance is largely reduced by dividing the tracer power spectrum by that from a BAO-free simulation starting with the same phases. This permits us to study with unprecedented accuracy (better than 0.02% for dark matter and 0.07% for low-bias halos) small shifts of the pristine BAO wavenumbers towards larger k, and non-linear damping of BAO wiggles in the power spectrum of dark matter and halo populations in the redshift range z=0-1. For dark matter, we provide an accurate parametrization of the evolution of αas a function of the linear growth factor D(z). For halo samples, with bias ranging from 1.2 to 2.8, we measure a typical BAO shift of ~0.25%, observed in real-space, which does not show an appreciable evolution with redshift within the uncertainties. Moreover, we report a constant shift as a function of halo bias. We find a different evolution of the damping of the acoustic feature in all halo samples as compared to dark matter with haloes suffering less damping, and also find some weak dependence on bias. A larger BAO shift and damping is measured in redshift-space which can be well explained by linear theory due to redshift-space distortions. A clear modulation in phase with the acoustic scale is observed in the scale-dependent halo bias due to the presence of the baryonic acoustic oscillations.

astro-ph.CO↗