arXiv subjects

Masayuki Tanaka

Publications and source records attributed to Masayuki Tanaka.

At least 91 records · Page 5Linked to original sources

arXiv2020

Human Segmentation with Dynamic LiDAR Data

Consecutive LiDAR scans compose dynamic 3D sequences, which contain more abundant information than a single frame. Similar to the development history of image and video perception, dynamic 3D sequence perception starts to come into sight after inspiring research on static 3D data perception. This work proposes a spatio-temporal neural network for human segmentation with the dynamic LiDAR point clouds. It takes a sequence of depth images as input. It has a two-branch structure, i.e., the spatial segmentation branch and the temporal velocity estimation branch. The velocity estimation branch is designed to capture motion cues from the input sequence and then propagates them to the other branch. So that the segmentation branch segments humans according to both spatial and temporal features. These two branches are jointly learned on a generated dynamic point cloud dataset for human recognition. Our works fill in the blank of dynamic point cloud perception with the spherical representation of point cloud and achieves high accuracy. The experiments indicate that the introduction of temporal feature benefits the segmentation of dynamic point cloud.

cs.CV↗

arXiv2020

Statistical correlation between the distribution of Ly$α$ emitters and IGM HI at $z\sim2.2$ mapped by Subaru/Hyper Suprime-Cam

The correlation between neutral Hydrogen (HI) in the intergalactic medium (IGM) and galaxies now attracts great interests. We select four fields which include several coherently strong Ly$α$ absorption systems at $z\sim2.2$ detected by using background quasars from the whole SDSS/(e)BOSS database. Deep narrow-band and $g$-band imaging are performed using the Hyper Suprime-Cam on the Subaru Telescope. We select out 2,642 Ly$α$ emitter (LAE) candidates at $z=2.177\pm0.023$ down to the Ly$α$ luminosity of $L_{\text{Ly}α}\approx 2 \times 10^{42} {\rm erg~s}^{-1}$ to construct the galaxy overdensity maps, covering an effective area of 5.39 deg$^2$. Combining the sample with the Ly$α$ absorption estimated from 64 (e)BOSS quasar spectra, we find a moderate to strong correlation between the LAE overdensity $δ_{\rm LAE}$ and the effective optical depth $τ_{\rm LoS}$ in line-of-sights, with $P$-value$=0.09\%$ ($<0.01\%$) when the field that contains a significant quasar overdensity is in(ex)cluded. The cross-correlation analysis also clearly suggests that up to $4\pm1$ pMpc, LAEs tend to cluster in the regions rich in HI gas, indicated by the high $τ_{\rm LoS}$, and avoid the low $τ_{\rm LoS}$ region where the HI gas is deficient. By averaging the $τ_{\rm LoS}$ as a function of the projected distance ($d$) to LAEs, we find a $30\%$ excess signal at $2σ$ level at $d<200$ pkpc, indicating the dense HI in circumgalactic medium, and a tentative excess at $400<d<600$ pkpc in IGM regime, corroborating the cross-correlation signal detected at about $0.5$ pMpc. These statistical analyses indicate that galaxy$-$IGM HI correlations exist on scales ranging from several hundred pkpc to several pMpc at $z\sim2.2$.

astro-ph.GA↗

arXiv2020

Monochrome and Color Polarization Demosaicking Using Edge-Aware Residual Interpolation

A division-of-focal-plane or microgrid image polarimeter enables us to acquire a set of polarization images in one shot. Since the polarimeter consists of an image sensor equipped with a monochrome or color polarization filter array (MPFA or CPFA), the demosaicking process to interpolate missing pixel values plays a crucial role in obtaining high-quality polarization images. In this paper, we propose a novel MPFA demosaicking method based on edge-aware residual interpolation (EARI) and also extend it to CPFA demosaicking. The key of EARI is a new edge detector for generating an effective guide image used to interpolate the missing pixel values. We also present a newly constructed full color-polarization image dataset captured using a 3-CCD camera and a rotating polarizer. Using the dataset, we experimentally demonstrate that our EARI-based method outperforms existing methods in MPFA and CPFA demosaicking.

eess.IV↗

arXiv2020

Unsupervised Learning of Image Segmentation Based on Differentiable Feature Clustering

The usage of convolutional neural networks (CNNs) for unsupervised image segmentation was investigated in this study. In the proposed approach, label prediction and network parameter learning are alternately iterated to meet the following criteria: (a) pixels of similar features should be assigned the same label, (b) spatially continuous pixels should be assigned the same label, and (c) the number of unique labels should be large. Although these criteria are incompatible, the proposed approach minimizes the combination of similarity loss and spatial continuity loss to find a plausible solution of label assignment that balances the aforementioned criteria well. The contributions of this study are four-fold. First, we propose a novel end-to-end network of unsupervised image segmentation that consists of normalization and an argmax function for differentiable clustering. Second, we introduce a spatial continuity loss function that mitigates the limitations of fixed segment boundaries possessed by previous work. Third, we present an extension of the proposed method for segmentation with scribbles as user input, which showed better accuracy than existing methods while maintaining efficiency. Finally, we introduce another extension of the proposed method: unseen image segmentation by using networks pre-trained with a few reference images without re-training the networks. The effectiveness of the proposed approach was examined on several benchmark datasets of image segmentation.

cs.CV↗

arXiv2020

A 16 deg$^2$ survey of emission-line galaxies at z<1.6 from HSC-SSP PDR2 and CHORUS

We have conducted a comprehensive survey of emission-line galaxies at $z\lesssim1.6$ based on narrowband (NB) imaging data taken with Hyper Suprime-Cam (HSC) on the Subaru telescope. In this paper, we update the catalogs of H$α$, [OIII], and [OII] emission-line galaxies using the data from the second Public Data Release (PDR2) of Subaru Strategic Program (SSP) of HSC and Cosmic HydrOgen Reionization Unveiled with Subaru (CHORUS) survey along with the spectroscopic redshifts for 2,019 emission-line galaxies selected with the PDR1 data. The wider effective coverage of NB816 and NB921, 16.3 deg$^2$ and 16.9 deg$^2$ respectively, are available in the Deep and UltraDeep layers of HSC-SSP from the PDR2. The CHORUS survey provides us with data with additional three NBs (NB527, NB718, and NB973) in the COSMOS field in the UltraDeep layer (1.37 deg$^2$). The five NB datasets allow us to investigate the star-forming galaxies presenting emission-lines at 14 specific redshifts ranging from $z\sim1.6$ down to $z\sim0.05$. We revisit the distribution of large-scale structures and luminosity functions (LFs) for the emission-line galaxies with the large samples of 75,377 emission-line galaxies selected. The redshift revolution of LFs shows that the star formation rate densities (SFRDs) decreases monotonically from $z\sim1.6$, which is consistent with the cosmic SFRD ever known. Our samples of emission-line galaxies covering a sufficiently large survey volume are useful to investigate the evolution of star-forming galaxies since the cosmic noon in a wide range of environments including galaxy clusters, filaments, and voids.

astro-ph.GA↗

arXiv2020

The UV Luminosity Function of Protocluster Galaxies at $z\sim4$: the Bright-end Excess and the Enhanced Star Formation Rate Density

We report the rest-frame ultraviolet luminosity function of $g$-dropout galaxies in 177 protocluster candidates (PC UVLF) at $z\sim4$ selected in the Hyper Suprime-Cam Subaru Strategic Program. Comparing with the UVLF of field galaxies at the same redshift, we find that the PC UVLF shows a significant excess towards the bright-end. This excess can not be explained by the contribution of only active galactic nuclei, and we also find that this is more significant in higher dense regions. Assuming that all protocluster members are located on the star formation main sequence, the PC UVLF can be converted into a stellar mass function. Consequently, our protocluster members are inferred to have a 2.8 times more massive characteristic stellar mass than that of the field Lyman break galaxies at the same redshift. This study, for the first time, clearly shows that the enhancement in star formation or stellar mass in overdense regions can generally be seen as early as at $z\sim4$. We also estimate the star formation rate density (SFRD) in protocluster regions as $\simeq 6-20\%$ of the cosmic SFRD, based on the measured PC UVLF after correcting for the selection incompleteness in our protocluster sample. This high value suggests that protoclusters make a non-negligible contribution to the cosmic SFRD at $z\sim4$, as previously suggested by simulations. Our results suggest that protoclusters are essential components for the galaxy evolution at $z\sim4$.

astro-ph.GA↗

arXiv2020

Classifying degraded images over various levels of degradation

Classification for degraded images having various levels of degradation is very important in practical applications. This paper proposes a convolutional neural network to classify degraded images by using a restoration network and an ensemble learning. The results demonstrate that the proposed network can classify degraded images over various levels of degradation well. This paper also reveals how the image-quality of training data for a classification network affects the classification performance of degraded images.

cs.CV↗

arXiv2020

Wide and Deep Exploration of Radio Galaxies with Subaru HSC (WERGS). III. Discovery of a z = 4.72 Radio Galaxy with Lyman Break Technique

We report a discovery of $z = 4.72$ radio galaxy, HSC J083913.17+011308.1, by using the Lyman break technique with the Hyper Suprime-Cam Subaru Strategic Survey (HSC-SSP) catalog for VLA FIRST radio sources. The number of known high-$z$ radio galaxies (HzRGs) at $z > 3$ is quite small to constrain the evolution of HzRGs so far. The deep and wide-area optical survey by HSC-SSP enables us to apply the Lyman break technique to a large search for HzRGs. For an HzRG candidate among pre-selected $r$-band dropouts with a radio detection, a follow-up optical spectroscopy with GMOS/Gemini has been performed. The obtained spectrum presents a clear Ly$α$ emission line redshifted to $z=4.72$. The SED fitting analysis with the rest-frame UV and optical photometries suggests the massive nature of this HzRG with $\log{M_*/M_{\odot}} = 11.4$. The small equivalent width of Ly$α$ and the moderately red UV colors indicate its dusty host galaxy, implying a chemically evolved and dusty system. The radio spectral index does not meet a criterion for an ultra-steep spectrum: $α^{325}_{1400}$ of $-1.1$ and $α^{150}_{1400}$ of $-0.9$, demonstrating that the HSC-SSP survey compensates for a sub-population of HzRGs which are missed in surveys focusing on an ultra-steep spectral index.

astro-ph.GA↗

arXiv2020

Stellar Velocity Dispersion of a Massive Quenching Galaxy at z=4.01

We present the first stellar velocity dispersion measurement of a massive quenching galaxy at z=4.01. The galaxy is first identified as a massive z>~4 galaxy with suppressed star formation from photometric redshifts based on deep multi-band data in the UKIDSS Ultra Deep Survey field. A follow-up spectroscopic observation with MOSFIRE on Keck revealed strong multiple absorption features, which are identified as Balmer absorption lines, giving a secure redshift of z=4.01. Thanks to the high S/N of the spectrum, we are able to estimate the stellar velocity dispersion, sigma=268+/-59 km/s. This velocity dispersion is consistent with that of massive galaxies today, implying no significant evolution in stellar velocity dispersion over the last 12 Gyr. Based on an upper limit on its physical size from deep optical images (r_eff<1.3 kpc), we find that its dynamical mass is consistent with the stellar mass inferred from photometry. Furthermore, the galaxy is located on the mass fundamental plane extrapolated from lower redshift galaxies. Combining all these results, we find that the velocity dispersion does not significantly evolve with redshift, although the size and mass of massive quenched galaxies do. This suggests that the mass in the core of massive galaxies does not evolve significantly, while most of the mass growth occurs in the outskirts of the galaxies, which also increases the size. This picture is consistent with a two-phase formation scenario in which mass and size growth is due to accretion in the outskirts of galaxies via mergers.

astro-ph.GA↗

arXiv2020

Photometric Redshifts for the Hyper Suprime-Cam Subaru Strategic Program Data Release 2

We present a description of the second data release for the photometric redshift (photo-$z$) of the Subaru Strategic Program for the Hyper-Suprime Cam survey. Our photo-$z$ products for the entire area in the Data Release 2 are publicly available, and both our point estimate catalog products and full PDFs can be retrieved from the data release site, \url{https://hsc-release.mtk.nao.ac.jp/}.

astro-ph.GA↗

arXiv2020

Learning-Based Human Segmentation and Velocity Estimation Using Automatic Labeled LiDAR Sequence for Training

In this paper, we propose an automatic labeled sequential data generation pipeline for human segmentation and velocity estimation with point clouds. Considering the impact of deep neural networks, state-of-the-art network architectures have been proposed for human recognition using point clouds captured by Light Detection and Ranging (LiDAR). However, one disadvantage is that legacy datasets may only cover the image domain without providing important label information and this limitation has disturbed the progress of research to date. Therefore, we develop an automatic labeled sequential data generation pipeline, in which we can control any parameter or data generation environment with pixel-wise and per-frame ground truth segmentation and pixel-wise velocity information for human recognition. Our approach uses a precise human model and reproduces a precise motion to generate realistic artificial data. We present more than 7K video sequences which consist of 32 frames generated by the proposed pipeline. With the proposed sequence generator, we confirm that human segmentation performance is improved when using the video domain compared to when using the image domain. We also evaluate our data by comparing with data generated under different conditions. In addition, we estimate pedestrian velocity with LiDAR by only utilizing data generated by the proposed pipeline.

cs.CV↗

arXiv2020

Block-wise Scrambled Image Recognition Using Adaptation Network

In this study, a perceptually hidden object-recognition method is investigated to generate secure images recognizable by humans but not machines. Hence, both the perceptual information hiding and the corresponding object recognition methods should be developed. Block-wise image scrambling is introduced to hide perceptual information from a third party. In addition, an adaptation network is proposed to recognize those scrambled images. Experimental comparisons conducted using CIFAR datasets demonstrated that the proposed adaptation network performed well in incorporating simple perceptual information hiding into DNN-based image classification.

cs.CV↗

arXiv2019

Balmer Break Galaxy Candidates at $z \sim 6$: a Potential View on the Star-Formation Activity at $z \gtrsim 14$

We search for galaxies with a strong Balmer break (Balmer Break Galaxies; BBGs) at $z \sim 6$ over a 0.41 deg$^2$ effective area in the COSMOS field. Based on rich imaging data, including data obtained with the Atacama Large Millimeter/submillimeter Array (ALMA), three candidates are identified by their extremely red $K - [3.6]$ colors as well as by non-detection in X-ray, optical, far-infrared (FIR), and radio bands. The non-detection in the deep ALMA observations suggests that they are not dusty galaxies but BBGs at $z \sim 6$, although contamination from Active Galactic Nuclei (AGNs) at $z \sim 0$ cannot be completely ruled out for the moment. Our spectral energy distribution (SED) analyses reveal that the BBG candidates at $z \sim 6$ have stellar masses of $\approx 5 \times 10^{10} M_{\odot}$ dominated by old stellar populations with ages of $\gtrsim 700$ Myr. Assuming that all the three candidates are real BBGs at $z \sim 6$, we estimate the stellar mass density (SMD) to be $2.4^{+2.3}_{-1.3} \times 10^{4} M_{\odot}$ Mpc$^{-3}$. This is consistent with an extrapolation from the lower redshift measurements. The onset of star formation in the three BBG candidates is expected to be several hundred million years before the observed epoch of $z \sim 6$. We estimate the star-formation rate density (SFRD) contributed by progenitors of the BBGs to be 2.4 -- 12 $\times 10^{-5} M_{\odot}$ yr$^{-1} $Mpc$^{-3}$ at $z > 14$ (99.7\% confidence range). Our result suggests a smooth evolution of the SFRD beyond $z = 8$.

astro-ph.GA↗

arXiv2019

Quiescent galaxies 1.5 billion years after the Big Bang and their progenitors

We report two secure ($z=3.775, 4.012$) and one tentative ($z\approx3.767$) spectroscopic confirmations of massive and quiescent galaxies through $K$-band observations with Keck/MOSFIRE and VLT/X-Shooter. The stellar continuum emission, the absence of strong nebular emission lines and the lack of significant far-infrared detections confirm the passive nature of these objects, disfavoring the alternative solution of low-redshift dusty star-forming interlopers. We derive stellar masses of $\mathrm{log}(M_{\star}/M_\odot)\sim11$ and ongoing star formation rates placing these galaxies $\gtrsim 1-2$ dex below the main sequence at their redshifts. The adopted parametrization of the star formation history suggests that these sources experienced a strong ($\langle \rm SFR \rangle \sim 1200-3500\,M_\odot\,\mathrm{yr}^{-1}$) and short ($\sim 50$ Myr) burst of star formation, peaking $\sim 150-500$ Myr before the time of observation, all properties reminiscent of the characteristics of sub-millimeter galaxies (SMGs) at $z>4$. We investigate this connection by comparing the comoving number densities and the properties of these two populations. We find a fair agreement only with the deepest sub-mm surveys detecting not only the most extreme starbursts, but also more normal galaxies. We support these findings by further exploring the Illustris-TNG cosmological simulation, retrieving populations of both fully quenched massive galaxies at $z\sim3-4$ and SMGs at $z\sim4-5$, with number densities and properties in agreement with the observations at $z\sim3$, but in increasing tension at higher redshift. Nevertheless, as suggested by the observations, not all the progenitors of quiescent galaxies at these redshifts shine as bright SMGs in their past and, similarly, not all bright SMGs quench by $z\sim3$, both fractions depending on the threshold assumed to define the SMGs themselves.

astro-ph.GA↗

arXiv2019

The Subaru HSC Galaxy Clustering with Photometric Redshift I: Dark Halo Masses Versus Baryonic Properties of Galaxies at 0.3<z<1.4

We present the clustering properties of low-$z$ $(z\leq1.4)$ galaxies selected by the Hyper Suprime-Cam Subaru Strategic Program Wide layer over $145$ deg$^{2}$. The wide-field and multi-wavelength observation yields $5,064,770$ galaxies at $0.3\leq z\leq1.4$ with photometric redshifts and physical properties. This enables the accurate measurement of angular correlation functions and subsequent halo occupation distribution (HOD) analysis allows the connection between baryonic properties and dark halo properties. The fraction of less-massive satellite galaxies at $z\lesssim1$ is found to be almost constant at $\sim20\%$, but it gradually decreases beyond $M_{\star} \sim 10^{10.4}h^{-2}M_{\odot}$. However, the abundance of satellite galaxies at $z>1$ is quite small even for less-massive galaxies due to the rarity of massive centrals at high-$z$. This decreasing trend is connected to the small satellite fraction of Lyman break galaxies at $z>3$. The stellar-to-halo mass ratios at $0.3\leq z\leq1.4$ are almost consistent with the predictions obtained using the latest empirical model; however, we identify small excesses from the theoretical model at the massive end. The pivot halo mass is found to be unchanged at $10^{11.9-12.1}h^{-1}M_{\odot}$ at $0.3\leq z\leq1.4$, and we systematically show that $10^{12}h^{-1}M_{\odot}$ is a universal pivot halo mass up to $z\sim5$ that is derived using only the clustering/HOD analyses. Nevertheless, halo masses with peaked instantaneous baryon conversion efficiencies are much smaller than the pivot halo mass regardless of a redshift, and the most efficient stellar-mass assembly is thought to be in progress in $10^{11.0-11.5}h^{-1}M_{\odot}$ dark haloes.

astro-ph.GA↗

arXiv2019

X-ray study of the double source plane gravitational lens system Eye of Horus observed with XMM-Newton

A double source plane (DSP) system is a precious probe for the density profile of distant galaxies and cosmological parameters. However, these measurements could be affected by the surrounding environment of the lens galaxy. Thus, it is important to evaluate the cluster-scale mass for detailed mass modeling. We observed the {\it Eye of Horus}, a DSP system discovered by the Subaru HSC--SSP, with XMM--Newton. We detected two X-ray extended emissions, originating from two clusters, one centered at the {\it Eye of Horus}, and the other located $\sim100$ arcsec northeast to the {\it Eye of Horus}. We determined the dynamical mass assuming hydrostatic equilibrium, and evaluated their contributions to the lens mass interior of the Einstein radius. The contribution of the former cluster is $1.1^{+1.2}_{-0.5}\times10^{12}~M_{\odot}$, which is $21-76\%$ of the total mass within the Einstein radius. The discrepancy is likely due to the complex gravitational structure along the line of sight. On the other hand, the contribution of the latter cluster is only $\sim2\%$ on the {\it Eye of Horus}. Therefore, the influence associated with this cluster can be ignored.

astro-ph.GA↗

arXiv2019

Evidence for the Cross-correlation between Cosmic Microwave Background Polarization Lensing from POLARBEAR and Cosmic Shear from Subaru Hyper Suprime-Cam

We present the first measurement of cross-correlation between the lensing potential, reconstructed from cosmic microwave background (CMB) {\it polarization} data, and the cosmic shear field from galaxy shapes. This measurement is made using data from the POLARBEAR CMB experiment and the Subaru Hyper Suprime-Cam (HSC) survey. By analyzing an 11~deg$^2$ overlapping region, we reject the null hypothesis at 3.5$σ$\ and constrain the amplitude of the {\bf cross-spectrum} to $\widehat{A}_{\rm lens}=1.70\pm 0.48$, where $\widehat{A}_{\rm lens}$ is the amplitude normalized with respect to the Planck~2018{} prediction, based on the flat $Λ$ cold dark matter cosmology. The first measurement of this {\bf cross-spectrum} without relying on CMB temperature measurements is possible due to the deep POLARBEAR map with a noise level of ${\sim}$6\,$μ$K-arcmin, as well as the deep HSC data with a high galaxy number density of $n_g=23\,{\rm arcmin^{-2}}$. We present a detailed study of the systematics budget to show that residual systematics in our results are negligibly small, which demonstrates the future potential of this cross-correlation technique.

astro-ph.CO↗

arXiv2019

A rapidly declining transient discovered with Subaru/Hyper Suprime-Cam

We perform a high-cadence transient survey with Subaru Hyper Suprime-Cam (HSC), which we call the Subaru HSC survey Optimized for Optical Transients (SHOOT). We conduct HSC imaging observations with time intervals of about one hour on two successive nights, and spectroscopic and photometric follow-up observations. A rapidly declining blue transient SHOOT14di at $z=0.4229$ is found in observations on two successive nights with an image subtraction technique. The rate of brightness change is $+1.28^{+0.40}_{-0.27}~{\rm mag~day^{-1}}$ ($+1.83^{+0.57}_{-0.39}~{\rm mag~day^{-1}}$) in the observer (rest) frame and the rest-frame color between $3400$ and $4400~\unicode[.8,0]{x212B}$ is $M_{\rm 3400\unicode[.8,0]{x212B}}-M_{\rm 4400\unicode[.8,0]{x212B}}=-0.4$. The nature of the object is investigated by comparing its peak luminosity, decline rate, and color with those of transients and variables previously observed, and those of theoretical models. None of the transients or variables share the same properties as SHOOT14di. Comparisons with theoretical models demonstrate that, while the emission from the cooling envelope of a Type IIb supernova shows a slower decline rate than SHOOT14di, and the explosion of a red supergiant star with a dense circumstellar wind shows a redder color than SHOOT14di, the shock breakout at the stellar surface of the explosion of a $25M_{\odot}$ red supergiant star with a small explosion energy of $\leq0.4\times10^{51}$ erg reproduces the multicolor light curve of SHOOT14di. This discovery shows that a high-cadence, multicolor optical transient survey at intervals of about one hour, and continuous and immediate follow-up observations, is important for studies of normal core-collapse supernovae at high redshifts.

astro-ph.HE↗