arXiv subjects

Christian Wolf

Publications and source records attributed to Christian Wolf.

At least 127 records · Page 7Linked to original sources

arXiv2018

A topological classification of locally constant potentials via zero-temperature measures

We provide a topological classification of locally constant functions over subshifts of finite type via their zero-temperature measures. Our approach is to analyze the relationship between the distribution of the zero-temperature measures and the boundary of higher dimensional generalized rotation sets. We also discuss the regularity of the localized entropy function on the boundary of the generalized rotation sets.

math.DS↗

arXiv2018

Object Level Visual Reasoning in Videos

Human activity recognition is typically addressed by detecting key concepts like global and local motion, features related to object classes present in the scene, as well as features related to the global context. The next open challenges in activity recognition require a level of understanding that pushes beyond this and call for models with capabilities for fine distinction and detailed comprehension of interactions between actors and objects in a scene. We propose a model capable of learning to reason about semantically meaningful spatiotemporal interactions in videos. The key to our approach is a choice of performing this reasoning at the object level through the integration of state of the art object detection networks. This allows the model to learn detailed spatial interactions that exist at a semantic, object-interaction relevant level. We evaluate our method on three standard datasets (Twenty-BN Something-Something, VLOG and EPIC Kitchens) and achieve state of the art results on all of them. Finally, we show visualizations of the interactions learned by the model, which illustrate object classes and their interactions corresponding to different activity classes.

cs.CV↗

arXiv2018

Computability at zero temperature

In this paper, we investigate the computability of thermodynamic invariants at zero temperature for one-dimensional subshifts of finite type. In particular, we prove that the residual entropy (i.e., the joint ground state entropy) is an upper semi-computable function on the space of continuous potentials, but it is not computable. Next, we consider locally constant potentials for which the zero-temperature measure is known to exist. We characterize the computability of the zero-temperature measure and its entropy for potentials that are constant on cylinders of a given length k. In particular, we show the existence of an open and dense set of locally constant potentials for which the zero-temperature measure can be computationally identified as an elementary periodic point measure. Finally, we show that our methods do not generalize to treat the case when k is not given

math.DS↗

arXiv2018

Glimpse Clouds: Human Activity Recognition from Unstructured Feature Points

We propose a method for human activity recognition from RGB data that does not rely on any pose information during test time and does not explicitly calculate pose information internally. Instead, a visual attention module learns to predict glimpse sequences in each frame. These glimpses correspond to interest points in the scene that are relevant to the classified activities. No spatial coherence is forced on the glimpse locations, which gives the module liberty to explore different points at each frame and better optimize the process of scrutinizing visual information. Tracking and sequentially integrating this kind of unstructured data is a challenge, which we address by separating the set of glimpses from a set of recurrent tracking/recognition workers. These workers receive glimpses, jointly performing subsequent motion tracking and activity prediction. The glimpses are soft-assigned to the workers, optimizing coherence of the assignments in space, time and feature space using an external memory module. No hard decisions are taken, i.e. each glimpse point is assigned to all existing workers, albeit with different importance. Our methods outperform state-of-the-art methods on the largest human activity recognition dataset available to-date; NTU RGB+D Dataset, and on a smaller human action recognition dataset Northwestern-UCLA Multiview Action 3D Dataset. Our code is publicly available at https://github.com/fabienbaradel/glimpse_clouds.

cs.CV↗

arXiv2018

OMEGA - OSIRIS Mapping of Emission-line Galaxies in A901/2: IV. - Extinction of Star-Formation Estimators with Inclination

We study the effect of inclination on the apparent brightness of star-forming galaxies in spectral passbands that are commonly used as star-formation indicators. As diagnostics we use mass-to-light ratios in three passbands: the UV continuum at 280 nm, the H$α$ emission line, and the FIR 24$μ$-band. We include a study of inclination trends in the IR/UV ratio ("IRX") and the IR/H$α$ ratio. Our sample comprises a few hundred galaxies from the region around the clusters Abell 901/902 with deep data and inclinations measured from outer disks in Hubble Space Telescope images. As a novelty, the H$α$- and separately the NII-emission are measured by tunable-filter imaging and encompass galaxies in their entirety. At galaxy stellar masses above log $M_*/M_\odot > 10$ we find trends in the UV and H$α$ mass-to-light ratio that suggest an inclination-induced attenuation from face-on to edge-on of $\sim 1$ mag and $\sim 0.7$ mag in UV and H$α$, respectively, implying that star-formation rates of edge-on galaxies would be underestimated by $\sim 2.5\times$ in UV and $\sim 2\times$ in H$α$. We find the luminosities in UV and H$α$ to be well correlated, but the optical depth of diffuse dust that causes inclination dependence appears to be lower for stars emitting at 280 nm than for gas clouds emitting Balmer lines. For galaxies with log $M_*/M_\odot < 9.7$, we find no measurable effect at $>0.1$ mag. The absence of an inclination dependence at 24$μ$ confirms that the average galaxy is optically thin in the FIR.

astro-ph.GA↗

arXiv2018

Discovery of the most ultra-luminous QSO using Gaia, SkyMapper and WISE

We report the discovery of the ultra-luminous QSO SMSS~J215728.21-360215.1 with magnitude $z=16.9$ and W4$=7.42$ at redshift 4.75. Given absolute magnitudes of $M_{145,\rm AB}=-29.3$, $M_{300,\rm AB}=-30.12$ and $\log L_{\rm bol}/L_{\rm bol,\odot} = 14.84$, it is the QSO with the highest unlensed UV-optical luminosity currently known in the Universe. It was found by combining proper-motion data from Gaia DR2 with photometry from SkyMapper DR1 and the Wide-field Infrared Survey Explorer (WISE). In the Gaia database it is an isolated single source and thus unlikely to be strongly gravitationally lensed. It is also unlikely to be a beamed source as it is not discovered in the radio domain by either NVSS or SUMSS. It is classed as a weak-emission-line QSO and possesses broad absorption line features. A lightcurve from ATLAS spanning the time from October 2015 to December 2017 shows little sign of variability.

astro-ph.GA↗

arXiv2018

Discovery of two bright $z\sim5$ quasars with SkyMapper, Pan-STARRS1 and WISE

We present a search for bright $z\sim5$ quasars using imaging data from SkyMapper Southern Survey, Pan-STARRS1 and the Wide-field Infrared Survey Explorer (WISE). We select two sets of candidates using WISE with optical bands from SkyMapper and alternatively from Pan-STARRS1, limited to a magnitude of $i<18.2$. We follow up several candidates with spectroscopy and find that the four candidates common to both lists are quasars, while others turned out to be cool stars. Two of the four quasars, SMSS J013539.27-212628.4 at $z=4.86$ and SMSS J093032.58-221207.7 at $z=4.94$, are new discoveries and ranked among the dozen brightest known $z>4.5$ QSOs in the $i$-band.

astro-ph.GA↗

arXiv2018

The Taipan Galaxy Survey: Scientific Goals and Observing Strategy

Taipan is a multi-object spectroscopic galaxy survey starting in 2017 that will cover 2pi steradians over the southern sky, and obtain optical spectra for about two million galaxies out to z<0.4. Taipan will use the newly-refurbished 1.2m UK Schmidt Telescope at Siding Spring Observatory with the new TAIPAN instrument, which includes an innovative 'Starbugs' positioning system capable of rapidly and simultaneously deploying up to 150 spectroscopic fibres (and up to 300 with a proposed upgrade) over the 6-deg diameter focal plane, and a purpose-built spectrograph operating from 370 to 870nm with resolving power R>2000. The main scientific goals of Taipan are: (i) to measure the distance scale of the Universe (primarily governed by the local expansion rate, H_0) to 1% precision, and the structure growth rate of structure to 5%; (ii) to make the most extensive map yet constructed of the mass distribution and motions in the local Universe, using peculiar velocities based on improved Fundamental Plane distances, which will enable sensitive tests of gravitational physics; and (iii) to deliver a legacy sample of low-redshift galaxies as a unique laboratory for studying galaxy evolution as a function of mass and environment. The final survey, which will be completed within 5 years, will consist of a complete magnitude-limited sample (i<17) of about 1.2x10^6 galaxies, supplemented by an extension to higher redshifts and fainter magnitudes (i<18.1) of a luminous red galaxy sample of about 0.8x10^6 galaxies. Observations and data processing will be carried out remotely and in a fully-automated way, using a purpose-built automated 'virtual observer' software and an automated data reduction pipeline. The Taipan survey is deliberately designed to maximise its legacy value, by complementing and enhancing current and planned surveys of the southern sky at wavelengths from the optical to the radio.

astro-ph.GA↗

arXiv2018

A shift map with a discontinuous entropy function

Let $f:X\to X$ be a continuous map on a compact metric space with finite topological entropy. Further, we assume that the entropy map $μ\mapsto h_μ(f)$ is upper semi-continuous. It is well-known that this implies the continuity of the localized entropy function of a given continuous potential $ϕ:X\to R$. In this note we show that this result does not carry over to the case of higher-dimensional potentials $Φ:X\to R^m$. Namely, we construct for a shift map $f$ a $2$-dimensional Lipschitz continuous potential $Φ$ with a discontinuous localized entropy function.

math.DS↗

arXiv2018

Learning to recognize touch gestures: recurrent vs. convolutional features and dynamic sampling

We propose a fully automatic method for learning gestures on big touch devices in a potentially multi-user context. The goal is to learn general models capable of adapting to different gestures, user styles and hardware variations (e.g. device sizes, sampling frequencies and regularities). Based on deep neural networks, our method features a novel dynamic sampling and temporal normalization component, transforming variable length gestures into fixed length representations while preserving finger/surface contact transitions, that is, the topology of the signal. This sequential representation is then processed with a convolutional model capable, unlike recurrent networks, of learning hierarchical representations with different levels of abstraction. To demonstrate the interest of the proposed method, we introduce a new touch gestures dataset with 6591 gestures performed by 27 people, which is, up to our knowledge, the first of its kind: a publicly available multi-touch gesture dataset for interaction. We also tested our method on a standard dataset of symbolic touch gesture recognition, the MMG dataset, outperforming the state of the art and reporting close to perfect performance.

cs.LG↗

arXiv2018

SkyMapper Southern Survey: First Data Release (DR1)

We present the first data release (DR1) of the SkyMapper Southern Survey, a hemispheric survey carried out with the SkyMapper Telescope at Siding Spring Observatory in Australia. Here, we present the survey strategy, data processing, catalogue construction and database schema. The DR1 dataset includes over 66,000 images from the Shallow Survey component, covering an area of 17,200 deg$^2$ in all six SkyMapper passbands $uvgriz$, while the full area covered by any passband exceeds 20,000 deg$^2$. The catalogues contain over 285 million unique astrophysical objects, complete to roughly 18 mag in all bands. We compare our $griz$ point-source photometry with PanSTARRS1 DR1 and note an RMS scatter of 2%. The internal reproducibility of SkyMapper photometry is on the order of 1%. Astrometric precision is better than 0.2 arcsec based on comparison with Gaia DR1. We describe the end-user database, through which data are presented to the world community, and provide some illustrative science queries.

astro-ph.IM↗

arXiv2017

Human Action Recognition: Pose-based Attention draws focus to Hands

We propose a new spatio-temporal attention based mechanism for human action recognition able to automatically attend to the hands most involved into the studied action and detect the most discriminative moments in an action. Attention is handled in a recurrent manner employing Recurrent Neural Network (RNN) and is fully-differentiable. In contrast to standard soft-attention based mechanisms, our approach does not use the hidden RNN state as input to the attention model. Instead, attention distributions are extracted using external information: human articulated pose. We performed an extensive ablation study to show the strengths of this approach and we particularly studied the conditioning aspect of the attention mechanism. We evaluate the method on the largest currently available human action recognition dataset, NTU-RGB+D, and report state-of-the-art results. Other advantages of our model are certain aspects of explanability, as the spatial and temporal attention distributions at test time allow to study and verify on which parts of the input data the method focuses.

cs.CV↗

arXiv2017

KiDS-450 + 2dFLenS: Cosmological parameter constraints from weak gravitational lensing tomography and overlapping redshift-space galaxy clustering

We perform a combined analysis of cosmic shear tomography, galaxy-galaxy lensing tomography, and redshift-space multipole power spectra (monopole and quadrupole) using 450 deg$^2$ of imaging data by the Kilo Degree Survey (KiDS) overlapping with two spectroscopic surveys: the 2-degree Field Lensing Survey (2dFLenS) and the Baryon Oscillation Spectroscopic Survey (BOSS). We restrict the galaxy-galaxy lensing and multipole power spectrum measurements to the overlapping regions with KiDS, and self-consistently compute the full covariance between the different observables using a large suite of $N$-body simulations. We methodically analyze different combinations of the observables, finding that galaxy-galaxy lensing measurements are particularly useful in improving the constraint on the intrinsic alignment amplitude (by 30%, positive at $3.5σ$ in the fiducial data analysis), while the multipole power spectra are useful in tightening the constraints along the lensing degeneracy direction (e.g. factor of two stronger matter density constraint in the fiducial analysis). The fully combined constraint on $S_8 \equiv σ_8 \sqrt{Ω_{\rm m}/0.3} = 0.742 \pm 0.035$, which is an improvement by 20% compared to KiDS alone, corresponds to a $2.6σ$ discordance with Planck, and is not significantly affected by fitting to a more conservative set of scales. Given the tightening of the parameter space, we are unable to resolve the discordance with an extended cosmology that is simultaneously favored in a model selection sense, including the sum of neutrino masses, curvature, evolving dark energy, and modified gravity. The complementarity of our observables allows for constraints on modified gravity degrees of freedom that are not simultaneously bounded with either probe alone, and up to a factor of three improvement in the $S_8$ constraint in the extended cosmology compared to KiDS alone.

astro-ph.CO↗

arXiv2017

PSYM-WIDE: a survey for large-separation planetary-mass companions to late spectral type members of young moving groups

We present the results of a direct-imaging survey for very large separation ($>$100 au), companions around 95 nearby young K5-L5 stars and brown dwarfs. They are high-likelihood candidates or confirmed members of the young ($\lessapprox$150 Myr) $β$ Pictoris and AB Doradus moving groups (ABDMG) and the TW Hya, Tucana-Horologium, Columba, Carina, and Argus associations. Images in $i'$ and $z'$ filters were obtained with the Gemini Multi-Object Spectrograph (GMOS) on Gemini South to search for companions down to an apparent magnitude of $z'\sim$22-24 at separations $\gtrapprox$20" from the targets and in the remainder of the wide 5.5' $\times$ 5.5' GMOS field of view. This allowed us to probe the most distant region where planetary-mass companions could be gravitationally bound to the targets. This region was left largely unstudied by past high-contrast imaging surveys, which probed much closer-in separations. This survey led to the discovery of a planetary-mass (9-13 $\,M_{\rm{Jup}}$) companion at 2000 au from the M3V star GU Psc, a highly probable member of ABDMG. No other substellar companions were identified. These results allowed us to constrain the frequency of distant planetary-mass companions (5-13 $\,M_{\rm{Jup}}$) to 0.84$_{-0.66}^{+6.73}$% (95% confidence) at semimajor axes between 500 and 5000 au around young K5-L5 stars and brown dwarfs. This is consistent with other studies suggesting that gravitationally bound planetary-mass companions at wide separations from low-mass stars are relatively rare.

astro-ph.EP↗

arXiv2017

Multi-view pose estimation with mixtures-of-parts and adaptive viewpoint selection

We propose a new method for human pose estimation which leverages information from multiple views to impose a strong prior on articulated pose. The novelty of the method concerns the types of coherence modelled. Consistency is maximised over the different views through different terms modelling classical geometric information (coherence of the resulting poses) as well as appearance information which is modelled as latent variables in the global energy function. Moreover, adequacy of each view is assessed and their contributions are adjusted accordingly. Experiments on the HumanEva and UMPM datasets show that the proposed method significantly decreases the estimation error compared to single-view results.

cs.CV↗

arXiv2017

Hand Pose Estimation through Semi-Supervised and Weakly-Supervised Learning

We propose a method for hand pose estimation based on a deep regressor trained on two different kinds of input. Raw depth data is fused with an intermediate representation in the form of a segmentation of the hand into parts. This intermediate representation contains important topological information and provides useful cues for reasoning about joint locations. The mapping from raw depth to segmentation maps is learned in a semi/weakly-supervised way from two different datasets: (i) a synthetic dataset created through a rendering pipeline including densely labeled ground truth (pixelwise segmentations); and (ii) a dataset with real images for which ground truth joint positions are available, but not dense segmentations. Loss for training on real images is generated from a patch-wise restoration process, which aligns tentative segmentation maps with a large dictionary of synthetic poses. The underlying premise is that the domain shift between synthetic and real data is smaller in the intermediate representation, where labels carry geometric and topological meaning, than in the raw input domain. Experiments on the NYU dataset show that the proposed training method decreases error on joints over direct regression of joints from depth data by 15.7%.

cs.CV↗

arXiv2017

Pose-conditioned Spatio-Temporal Attention for Human Action Recognition

We address human action recognition from multi-modal video data involving articulated pose and RGB frames and propose a two-stream approach. The pose stream is processed with a convolutional model taking as input a 3D tensor holding data from a sub-sequence. A specific joint ordering, which respects the topology of the human body, ensures that different convolutional layers correspond to meaningful levels of abstraction. The raw RGB stream is handled by a spatio-temporal soft-attention mechanism conditioned on features from the pose network. An LSTM network receives input from a set of image locations at each instant. A trainable glimpse sensor extracts features on a set of predefined locations specified by the pose stream, namely the 4 hands of the two people involved in the activity. Appearance features give important cues on hand motion and on objects held in each hand. We show that it is of high interest to shift the attention to different hands at different time steps depending on the activity itself. Finally a temporal attention mechanism learns how to fuse LSTM features over time. We evaluate the method on 3 datasets. State-of-the-art results are achieved on the largest dataset for human activity recognition, namely NTU-RGB+D, as well as on the SBU Kinect Interaction dataset. Performance close to state-of-the-art is achieved on the smaller MSR Daily Activity 3D dataset.

cs.CV↗

arXiv2017

Residual Conv-Deconv Grid Network for Semantic Segmentation

This paper presents GridNet, a new Convolutional Neural Network (CNN) architecture for semantic image segmentation (full scene labelling). Classical neural networks are implemented as one stream from the input to the output with subsampling operators applied in the stream in order to reduce the feature maps size and to increase the receptive field for the final prediction. However, for semantic image segmentation, where the task consists in providing a semantic class to each pixel of an image, feature maps reduction is harmful because it leads to a resolution loss in the output prediction. To tackle this problem, our GridNet follows a grid pattern allowing multiple interconnected streams to work at different resolutions. We show that our network generalizes many well known networks such as conv-deconv, residual or U-Net networks. GridNet is trained from scratch and achieves competitive results on the Cityscapes dataset.

cs.CV↗