Search

Despite the success in 6D pose estimation in bin-picking scenarios, existing methods still struggle to produce accurate prediction results for symmetry objects and real world scenarios. The primary bottlenecks include 1) the ambiguity keypoints caused by object symmetries; 2) the domain gap between real and synthetic data. To circumvent these problem, we propose a new 6D pose estimation network with symmetric-aware keypoint prediction and self-training domain adaptation (SD-Net). SD-Net builds on pointwise keypoint regression and deep hough voting to perform reliable detection keypoint under clutter and occlusion. Specifically, at the keypoint prediction stage, we designe a robust 3D keypoints selection strategy considering the symmetry class of objects and equivalent keypoints, which facilitate locating 3D keypoints even in highly occluded scenes. Additionally, we build an effective filtering algorithm on predicted keypoint to dynamically eliminate multiple ambiguity and outlier keypoint candidates. At the domain adaptation stage, we propose the self-training framework using a student-teacher training scheme. To carefully distinguish reliable predictions, we harnesses a tailored heuristics for 3D geometry pseudo labelling based on semi-chamfer distance. On public Sil'eane dataset, SD-Net achieves state-of-the-art results, obtaining an average precision of 96%. Testing learning and generalization abilities on public Parametric datasets, SD-Net is 8% higher than the state-of-the-art method. The code is available at https://github.com/dingthuang/SD-Net.

Existing Object Pose Estimation (OPE) methods for stacked scenarios are not robust to changes in object scale. This paper proposes a new 6DoF OPE network (NormNet) for different scale objects in stacked scenarios. Specifically, each object's scale is first learned with point-wise regression. Then, all objects in the stacked scenario are normalized into the same scale through semantic segmentation and affine transformation. Finally, they are fed into a shared pose estimator to recover their 6D poses. In addition, we introduce a new Sim-to-Real transfer pipeline, combining style transfer and domain randomization. This improves the NormNet's performance on real data even if we only train it on synthetic data. Extensive experiments demonstrate that the proposed method achieves state-of-the-art performance on public benchmarks and the MultiScale dataset we constructed. The real-world experiments show that our method can robustly estimate the 6D pose of objects at different scales.

The discovery of the optical counterpart, along with the gravitational waves from GW170817, of the first binary neutron star merger, opened up a new era for multi-messenger astrophysics. Combining the GW data with the optical counterpart, also known as AT2017gfo, classified as a kilonova, has revealed the nature of compact binary merging systems by extracting enriched information about the total binary mass, the mass ratio, the system geometry, and the equation of state. Even though the detection of kilonova brought about a revolution in the domain of multi-messenger astronomy, since there has been only one kilonova from a gravitational wave detected binary neutron star merger event so far, this limits the exact understanding of the origin and propagation of the kilonova. Here, we use a conditional variational autoencoder trained on light curve data from two kilonova models having different temporal lengths, and consequently, generate kilonova light curves rapidly based on physical parameters of our choice with good accuracy. Once trained, the time scale for light curve generation is of the order of a few milliseconds, thus speeding up generating light curves by $1000$ times compared to the simulation. The mean squared error between the generated and original light curves is typically $0.015$ with a maximum of $0.08$ for each set of considered physical parameter; while having a maximum of $\approx0.6$ error across the whole parameter space. Hence, implementing this technique provides fast and reliably accurate results., Comment: 19 pages, 7 figures (3 additional figures in appendix), accepted to ApJ

Fulfilling the rich promise of rapid advances in time-domain astronomy is only possible through confronting our observations with physical models and extracting the parameters that best describe what we see. Here, we introduce {\sc Redback}; a Bayesian inference software package for electromagnetic transients. {\sc Redback} provides an object-orientated {\sc python} interface to over 12 different samplers and over 100 different models for kilonovae, supernovae, gamma-ray burst afterglows, tidal disruption events, engine-driven transients among other explosive transients. The models range in complexity from simple analytical and semi-analytical models to surrogates built upon numerical simulations accelerated via machine learning. {\sc Redback} also provides a simple interface for downloading and processing data from various catalogs such as \textit{Swift} and Fink. The software can also serve as an engine to simulate transients for telescopes such as the Zwicky Transient Facility and Vera Rubin with realistic cadences, limiting magnitudes, and sky-coverage or a hypothetical user-constructed survey or a generic transient for target-of-opportunity observations with different telescopes. As a demonstration of its capabilities, we show how {\sc Redback} can be used to jointly fit the spectrum and photometry of a kilonova, enabling a more powerful, holistic probe into the properties of a transient. We also showcase general examples of how {\sc Redback} can be used as a tool to simulate transients for realistic surveys, fit models to real, simulated, or private data, multi-messenger inference with gravitational waves, and serve as an end-to-end software toolkit for parameter estimation and interpreting the nature of electromagnetic transients., Comment: Published in MNRAS. 25 pages 11 figures. Redback is available on GitHub at https://github.com/nikhil-sarin/redback

We present a novel fully Bayesian analysis to constrain short gamma-ray burst jet structures associated with cocoon, wide-angle and simple top-hat jet models, as well as the binary neutron star merger rate. These constraints are made given the distance and inclination information from GW170817, observed flux of GRB170817A, observed rate of short gamma-ray bursts detected by Swift, and the neutron star merger rate inferred from LIGO's first and second observing runs. A separate analysis is conducted where a fitted short gamma-ray burst luminosity function is included to provide further constraints. The jet structure models are further constrained using the observation of GW190425 and we find that the assumption that it produced a GRB170817-like short gamma-ray burst that went undetected due to the jet geometry is consistent with previous observations. We find and quantify evidence for low luminosity and wide-angled jet structuring in the short gamma-ray burst population, independently from afterglow observations, with log Bayes factors of $0.45{-}0.55$ for such models when compared to a classical top-hat jet. Slight evidence is found for a Gaussian jet structure model over all others when the fitted luminosity function is provided, producing log Bayes factors of $0.25{-}0.9\pm0.05$ when compared to the other models. However without considering GW190425 or the fitted luminosity function, the evidence favours a cocoon-like model with log Bayes factors of $0.14\pm0.05$ over the Gaussian jet structure. We provide new constraints to the binary neutron star merger rates of $1{-}1300$Gpc$^{-3}$yr$^{-1}$ or $2{-}680$Gpc$^{-3}$yr$^{-1}$ when a fitted luminosity function is assumed., Comment: 18 pages, 8 figures

Observing nearby galaxies with submillimeter telescopes on the ground has two major challenges. First, the brightness is significantly reduced at long submillimeter wavelengths compared to the brightness at the peak of the dust emission. Second, it is necessary to use a high-pass spatial filter to remove atmospheric noise on large angular scales, which has the unwelcome by-product of also removing the galaxy's large-scale structure. We have developed a technique for producing high-resolution submillimeter images of galaxies of large angular size by using the telescope on the ground to determine the small-scale structure (the large Fourier components) and a space telescope (Herschel or Planck) to determine the large-scale structure (the small Fourier components). Using this technique, we are carrying out the HARP and SCUBA-2 High Resolution Terahertz Andromeda Galaxy Survey (HASHTAG), an international Large Program on the James Clerk Maxwell Telescope, with one aim being to produce the first high-fidelity high-resolution submillimeter images of Andromeda. In this paper, we describe the survey, the method we have developed for combining the space-based and ground-based data, and present the first HASHTAG images of Andromeda at 450 and 850um. We also have created a method to predict the CO(J=3-2) line flux across M31, which contaminates the 850um band. We find that while normally the contamination is below our sensitivity limit, the contamination can be significant (up to 28%) in a few of the brightest regions of the 10 kpc ring. We therefore also provide images with the predicted line emission removed., Comment: 26 pages, 19 figures. Submitted to ApJS June 2021, Accepted September 2021

In the field of multi-messenger astronomy, Bayesian inference is commonly adopted to compare the compatibility of models given the observed data. However, to describe a physical system like neutron star mergers and their associated gamma-ray burst (GRB) events, usually more than ten physical parameters are incorporated in the model. With such a complex model, likelihood evaluation for each Monte Carlo sampling point becomes a massive task and requires a significant amount of computational power. In this work, we perform quick parameter estimation on simulated GRB X-ray light curves using an interpolated physical GRB model. This is achieved by generating a grid of GRB afterglow light curves across the parameter space and replacing the likelihood with a simple interpolation function in the high-dimensional grid that stores all light curves. This framework, compared to the original method, leads to a $\sim$90$\times$ speedup per likelihood estimation. It will allow us to explore different jet models and enable fast model comparison in the future., Comment: 9 pages, 4 figures, accepted to the special issue of Universe, "Waiting for GODOT -- Present and Future of Multi-Messenger Astronomy"

For gravitational wave (GW) detected neutron star mergers, one of the leading candidates for electromagnetic (EM) counterparts is the afterglow from an ultra-relativistic jet. Where this afterglow is observed, it will likely be viewed off-axis, such as the afterglow following GW170817/GRB 170817A. The temporal behaviour of an off-axis observed GRB afterglow can be used to reveal the lateral jet structure, and statistical model fits can put constraints on the various model free-parameters. Amongst these parameters is the inclination of the system to the line of sight. Along with the GW detection, the afterglow modelling provides the best constraint on the inclination to the line-of-sight and can improve the estimates of cosmological parameters e.g. the Hubble constant, from GW-EM events. However, modelling of the afterglow depends on the assumed jet structure and, often overlooked, the effects of lateral spreading. Here we show how the inclusion of lateral spreading in the afterglow models can affect the estimated inclination of GW-EM events., Comment: 14 pages, 5 figures, Accepted to the special issue of Universe, "Waiting for GODOT -- Present and Future of Multi-Messenger Astronomy"

Recently, there has been a panoptic segmentation task combining semantic and instance segmentation, in which the goal is to classify each pixel with the corresponding instance ID. In this work, we propose a solution to tackle the panoptic segmentation task. The overall structure combines the bottom-up method and the top-down method. Therefore, not only can there be better performance, but also the execution speed can be maintained. The network mainly pays attention to the quality of the mask. In the previous work, we can see that the uneven contour of the object is more likely to appear, resulting in low-quality prediction. Accordingly, we propose enhancement features and corresponding loss functions for the silhouette of objects and backgrounds to improve the mask. Meanwhile, we use the new proposed confidence score to solve the occlusion problem and make the network tend to use higher quality masks as prediction results. To verify our research, we used the COCO dataset and CityScapes dataset to do experiments and obtained competitive results with fast inference time., Comment: Technical report

Efficient representation of patients is very important in the healthcare domain and can help with many tasks such as medical risk prediction. Many existing methods, such as diagnostic Cost Groups (DCG), rely on expert knowledge to build patient representation from medical data, which is resource consuming and non-scalable. Unsupervised machine learning algorithms are a good choice for automating the representation learning process. However, there is very little research focusing on onpatient-level representation learning directly from medical claims. In this paper, weproposed a novel patient vector learning architecture that learns high quality,fixed-length patient representation from claims data. We conducted several experiments to test the quality of our learned representation, and the empirical results show that our learned patient vectors are superior to vectors learned through other methods including a popular commercial model. Lastly, we provide potential clinical interpretation for using our representation on predictive tasks, as interpretability is vital in the healthcare domain

For the first time, we present a Bayesian time-resolved spectral study of the X-ray afterglow datasets of GW170817/GRB17017A observed by the Chandra X-ray Observatory. These include all 12 public datasets, from the earliest observation taken at $\rm t \sim 9~d$ to the newest observation at $\rm \sim 359~d$ post-merger. While our results are consistent with the other works using Cash statistic within uncertainty, the Bayesian analysis we performed in this work have yielded Gaussian-like parameter distributions. We also obtained the parameter uncertainties directly from their posterior probability distributions. We are able to confirm that the power-law photon index has remained constant of $\Gamma \sim 1.6$ throughout the entire year-long observing period, except for the first dataset observed at $\rm t = 8.9~d$ when $\Gamma =1.04\pm0.44$ is marginally harder. We also found that the unabsorbed X-ray flux peaked at $\rm t \sim 155~d$, temporally consistent with the X-ray flare model suggested recently by Piro et al (2018). The X-ray flux has been fading since $\sim160$ days after the merger and has returned to the level as first discovered after one year. Our result shows that the X-ray spectrum of GW170817/GRB170817A is well-described by a simple power-law originated from non-thermal slow-cooling synchrotron radiation., Comment: Accepted for publication in the Journal of High Energy Astrophysics