arXiv subjects

Michele Sasdelli

Publications and source records attributed to Michele Sasdelli.

26 records · Page 2Linked to original sources

arXiv2018

Improving the Annotation of DeepFashion Images for Fine-grained Attribute Recognition

DeepFashion is a widely used clothing dataset with 50 categories and more than overall 200k images where each image is annotated with fine-grained attributes. This dataset is often used for clothes recognition and although it provides comprehensive annotations, the attributes distribution is unbalanced and repetitive specially for training fine-grained attribute recognition models. In this work, we tailored DeepFashion for fine-grained attribute recognition task by focusing on each category separately. After selecting categories with sufficient number of images for training, we remove very scarce attributes and merge the duplicate ones in each category, then we clean the dataset based on the new list of attributes. We use a bilinear convolutional neural network with pairwise ranking loss function for multi-label fine-grained attribute recognition and show that the new annotations improve the results for such a task. The detailed annotations for each of the selected categories are provided for public use.

cs.CV↗

arXiv2018

Hide and Seek tracker: Real-time recovery from target loss

In this paper, we examine the real-time recovery of a video tracker from a target loss, using information that is already available from the original tracker and without a significant computational overhead. More specifically, before using the tracker output to update the target position we estimate the detection confidence. In the case of a low confidence, the position update is rejected and the tracker passes to a single-frame failure mode, during which the patch low-level visual content is used to swiftly update the object position, before recovering from the target loss in the next frame. Orthogonally to this improvement, we further enhance the running average method used for creating the query model in tracking-through-similarity. The experimental evidence provided by evaluation on standard tracking datasets (OTB-50, OTB-100 and OTB-2013) validate that target recovery can be successfully achieved without compromising the real-time update of the target position.

cs.CV↗

arXiv2016

A metric space for type Ia supernova spectra: a new method to assess explosion scenarios

Over the past years type Ia supernovae (SNe Ia) have become a major tool to determine the expansion history of the Universe, and considerable attention has been given to, both, observations and models of these events. However, until now, their progenitors are not known. The observed diversity of light curves and spectra seems to point at different progenitor channels and explosion mechanisms. Here, we present a new way to compare model predictions with observations in a systematic way. Our method is based on the construction of a metric space for SN Ia spectra by means of linear Principal Component Analysis (PCA), taking care of missing and/or noisy data, and making use of Partial Least Square regression (PLS) to find correlations between spectral properties and photometric data. We investigate realizations of the three major classes of explosion models that are presently discussed: delayed-detonation Chandrasekhar-mass explosions, sub-Chandrasekhar-mass detonations, and double-degenerate mergers, and compare them with data. We show that in the PC space all scenarios have observed counterparts, supporting the idea that different progenitors are likely. However, all classes of models face problems in reproducing the observed correlations between spectral properties and light curves and colors. Possible reasons are briefly discussed.

astro-ph.HE↗

arXiv2016

Exploring the spectroscopic diversity of type Ia supernovae with DRACULA: a machine learning approach

The existence of multiple subclasses of type Ia supernovae (SNeIa) has been the subject of great debate in the last decade. One major challenge inevitably met when trying to infer the existence of one or more subclasses is the time consuming, and subjective, process of subclass definition. In this work, we show how machine learning tools facilitate identification of subtypes of SNeIa through the establishment of a hierarchical group structure in the continuous space of spectral diversity formed by these objects. Using Deep Learning, we were capable of performing such identification in a 4 dimensional feature space (+1 for time evolution), while the standard Principal Component Analysis barely achieves similar results using 15 principal components. This is evidence that the progenitor system and the explosion mechanism can be described by a small number of initial physical parameters. As a proof of concept, we show that our results are in close agreement with a previously suggested classification scheme and that our proposed method can grasp the main spectral features behind the definition of such subtypes. This allows the confirmation of the velocity of lines as a first order effect in the determination of SNIa subtypes, followed by 91bg-like events. Given the expected data deluge in the forthcoming years, our proposed approach is essential to allow a quick and statistically coherent identification of SNeIa subtypes (and outliers). All tools used in this work were made publicly available in the Python package Dimensionality Reduction And Clustering for Unsupervised Learning in Astronomy (DRACULA) and can be found within COINtoolbox (https://github.com/COINtoolbox/DRACULA).

astro-ph.SR↗

arXiv2016

Luminosity distributions of Type Ia Supernovae

We have assembled a dataset of 165 low redshift, $z<$0.06, publicly available type Ia supernovae (SNe Ia). We produce maximum light magnitude ($M_{B}$ and $M_{V}$) distributions of SNe Ia to explore the diversity of parameter space that they can fill. Before correction for host galaxy extinction we find that the mean $M_{B}$ and $M_{V}$ of SNe Ia are $-18.58\pm0.07$mag and $-18.72\pm0.05$mag respectively. Host galaxy extinction is corrected using a new method based on the SN spectrum. After correction, the mean values of $M_{B}$ and $M_{V}$ of SNe Ia are $-19.10\pm0.06$ and $-19.10\pm0.05$mag respectively. After correction for host galaxy extinction, `normal' SNeIa ($Δm_{15}(B)<1.6$mag) fill a larger parameter space in the Width-Luminosity Relation (WLR) than previously suggested, and there is evidence for luminous SNe Ia with large $Δm_{15}(B)$. We find a bimodal distribution in $Δm_{15}(B)$, with a pronounced lack of transitional events at $Δm_{15}(B)$=1.6 mag. We confirm that faster, low-luminosity SNe tend to come from passive galaxies. Dividing the sample by host galaxy type, SNe Ia from star-forming (S-F) galaxies have a mean $M_{B}=-19.20 \pm 0.05$ mag, while SNe Ia from passive galaxies have a mean $M_{B}=-18.57 \pm 0.24$ mag. Even excluding fast declining SNe, `normal' ($M_{B}<-18$ mag) SNe Ia from S-F and passive galaxies are distinct. In the $V$-band, there is a difference of 0.4$ \pm $0.13 mag between the median ($M_{V}$) values of the `normal' SN Ia population from passive and S-F galaxies. This is consistent with ($\sim 15 \pm $10)% of `normal' SNe Ia from S-F galaxies coming from an old stellar population.

astro-ph.HE↗

arXiv2016

A Luminous Peculiar Type Ia Supernova SN 2011hr: More Like SN 1991T or SN 2007if?

Photometric and spectroscopic observations of a slowly declining, luminous Type Ia supernova (SN Ia) SN 2011hr in the starburst galaxy NGC 2691 are presented. SN 2011hr is found to peak at $M_{B}=-19.84 \pm 0.40\,\rm{mag}$, with a post-maximum decline rate $Δ$m$_{15}$(B) = 0.92 $\pm$ 0.03\,$\rm{mag}$. From the maximum-light bolometric luminosity, $L=(2.30 \pm 0.90) \times 10^{43}\,\rm{erg\,s^{-1}}$, we estimate the mass of synthesized \Nifs\ in SN 2011hr to be $M(\rm{^{56}Ni})=1.11 \pm 0.43\,M_{\sun}$. SN 2011hr appears more luminous than SN 1991T at around maximum light, and the absorption features from its intermediate-mass elements (IMEs) are noticeably weaker than the latter at similar phases. Spectral modeling suggests that SN 2011hr has the IMEs of $\sim$\,0.07 M$_{\sun}$ in the outer ejecta, which is much lower than the typical value of normal SNe Ia (i.e., 0.3 -- 0.4 M$_{\sun}$) and is also lower than the value of SN 1991T (i.e., $\sim$\,0.18 M$_{\sun}$). These results indicate that SN 2011hr may arise from a Chandrasekhar-mass white dwarf progenitor that experienced a more efficient burning process in the explosion. Nevertheless, it is still possible that SN 2011hr may serve as a transitional object connecting the SN 1991T-like SNe Ia with the superluminous subclass like SN 2007if given that the latter also shows very weak IMEs at all phases.

astro-ph.SR↗

arXiv2014

A metric space for type Ia supernova spectra

We develop a new framework for use in exploring Type Ia Supernova (SN Ia) spectra. Combining Principal Component Analysis (PCA) and Partial Least Square analysis (PLS) we are able to establish correlations between the Principal Components (PCs) and spectroscopic/photometric SNe Ia features. The technique was applied to ~120 supernova and ~800 spectra from the Nearby Supernova Factory. The ability of PCA to group together SNe Ia with similar spectral features, already explored in previous studies, is greatly enhanced by two important modifications: (1) the initial data matrix is built using derivatives of spectra over the wavelength, which increases the weight of weak lines and discards extinction, and (2) we extract time evolution information through the use of entire spectral sequences concatenated in each line of the input data matrix. These allow us to define a stable PC parameter space which can be used to characterize synthetic SN Ia spectra by means of real SN features. Using PLS, we demonstrate that the information from important previously known spectral indicators (namely the pseudo-equivalent width (pEW) of Si II 5972 / Si II 6355 and the line velocity of S II 5640 / Si II 6355) at a given epoch, is contained within the PC space and can be determined through a linear combination of the most important PCs. We also show that the PC space encompasses photometric features like B or V magnitudes, B-V color and SALT2 parameters c and x1. The observed colors and magnitudes, that are heavily affected by extinction, cannot be reconstructed using this technique alone. All the above mentioned applications allowed us to construct a metric space for comparing synthetic SN Ia spectra with observations.

astro-ph.SR↗

arXiv2014

Abundance stratification in Type Ia Supernovae - IV: the luminous, peculiar SN 1991T

The abundance distribution of the elements in the ejecta of the peculiar, luminous Type Ia supernova (SN Ia) 1991T is obtained modelling spectra from before maximum light until a year after the explosion, with the method of "Abundance Tomography". SN 1991T is different from other slowly declining SNe Ia (e.g. SN 1999ee) in having a weaker Si II 6355 line and strong features of iron group elements before maximum. The distance to the SN is investigated along with the abundances and the density profile. The ionization transition that happens around maximum sets a strict upper limit on the luminosity. Both W7 and the WDD3 delayed detonation model are tested. WDD3 is found to provide marginally better fits. In this model the core of the ejecta is dominated by stable Fe with a mass of about 0.15 solar masses, as in most SNe Ia. The layer above is mainly 56Ni up to v~10000 km/s (~0.78 solar masses). A significant amount of 56Ni (~3 %) is located in the outer layers. A narrow layer between 10000 km/s and ~12000 km/s is dominated by intermediate mass elements (IME), ~0.18 solar masses. This is small for a SN Ia. The high luminosity and the consequently high ionization, and the high 56Ni abundance at high velocities explain the peculiar early-time spectra of SN 1991T. The outer part is mainly of oxygen, ~0.3 solar masses. Carbon lines are never detected, yielding an upper limit of 0.01 solar masses for C. The abundances obtained with the W7 density model are qualitatively similar to those of the WDD3 model. Different elements are stratified with moderate mixing, resembling a delayed detonation.

astro-ph.SR↗