Humans can easily infer the underlying 3D geometry and texture of an object only from a single 2D image. Current computer vision methods can do this, too, but suffer from view generalization problems - the models inferred tend to make poor prediction
s of appearance in novel views. As for generalization problems in machine learning, the difficulty is balancing single-view accuracy (cf. training error; bias) with novel view accuracy (cf. test error; variance). We describe a class of models whose geometric rigidity is easily controlled to manage this tradeoff. We describe a cycle consistency loss that improves view generalization (roughly, a model from a generated view should predict the original view well). View generalization of textures requires that models share texture information, so a car seen from the back still has headlights because other cars have headlights. We describe a cycle consistency loss that encourages model textures to be aligned, so as to encourage sharing. We compare our method against the state-of-the-art method and show both qualitative and quantitative improvements.
High-resolution observations of ionized and molecular gas in the nuclear regions of galaxies are indispensable for delineating the interplay of star formation, gaseous inflows, stellar radiation, and feedback processes. Combining our new ALMA band 3
mapping and archival VLT/MUSE data, we present a spatially resolved analysis of molecular and ionized gas in the central 5.4 Kpc region of NGC 1365. We find the star formation rate/efficiency (SFR/SFE) in the inner circumnuclear ring is about 0.4/1.1 dex higher than in the outer regions. At a linear resolution of 180 pc, we obtain a super-linear Kennicutt-Schmidt law, demonstrating a steeper slope (1.96$pm$0.14) than previous results presumably based on lower-resolution observations. Compared to the northeastern counterpart, the southwestern dust lane shows lower SFE, but denser molecular gas, and larger virial parameters. This is consistent with an interpretation of negative feedback from AGN and/or starburst, in the sense that the radiation/winds can heat and interact with the molecular gas even in relatively dense regions. After subtracting the circular motion component of the molecular gas and the stellar rotation, we detect two prominent non-circular motion components of molecular and ionized hydrogen gas, reaching a line-of-sight velocity of up to 100 km/s. We conclude that the winds or shocked gas from the central AGN may expel the low-density molecular gas and diffuse ionized gas on the surface of the rotating disk.
Generative Adversarial Networks (GANs) produce impressive results on unconditional image generation when powered with large-scale image datasets. Yet generated images are still easy to spot especially on datasets with high variance (e.g. bedroom, chu
rch). In this paper, we propose various improvements to further push the boundaries in image generation. Specifically, we propose a novel dual contrastive loss and show that, with this loss, discriminator learns more generalized and distinguishable representations to incentivize generation. In addition, we revisit attention and extensively experiment with different attention blocks in the generator. We find attention to be still an important module for successful image generation even though it was not used in the recent state-of-the-art models. Lastly, we study different attention architectures in the discriminator, and propose a reference attention mechanism. By combining the strengths of these remedies, we improve the compelling state-of-the-art Fr{e}chet Inception Distance (FID) by at least 17.5% on several benchmark datasets. We obtain even more significant improvements on compositional synthetic scenes (up to 47.5% in FID).
We present an analysis of the variability of broad absorption lines (BALs) in a quasar SDSS J141955.26+522741.1 at $z=2.145$ with 72 observations from the Sloan Digital Sky Survey Data Release 16 (SDSS DR16). The strong correlation between the equiva
lent widths of BAL and the continuum luminosity, reveals that the variation of BAL trough is dominated by the photoionization. The photoionization model predicts that when the time interval $Delta T$ between two observations is longer than the recombination timescale $t_{rm rec}$, the BAL variations can be detected. This can be characterized as a sharp rise in the detection rate of BAL variation at $Delta T=t_{rm rec}$. For the first time, we detect such a sharp rise signature in the detection rate of BAL variations. As a result, we propose that the $t_{rm rec}$ can be obtained from the sharp rise of the detection rate of BAL variation. It is worth mentioning that the BAL variations are detected at the time-intervals less than the $t_{rm rec}$ for half an order of magnitude in two individual troughs. This result indicates that there may be multiple components with different $t_{rm rec}$ but the same velocity in an individual trough.
Recent hydrodynamic simulations and observations of radio jets have shown that the surrounding environment has a large effect on their resulting morphology. To investigate this we use a sample of 50 Extended Radio Active Galactic Nuclei (ERAGN) detec
ted in the Observations of Redshift Evolution in Large Scale Environments (ORELSE) survey. These sources are all successfully cross-identified to galaxies within a redshift range of $0.55 leq z leq 1.35$, either through spectroscopic redshifts or accurate photometric redshifts. We find that ERAGN are more compact in high-density environments than those in low-density environments at a significance level of 4.5$sigma$. Among a series of internal properties under our scrutiny, only the radio power demonstrates a positive correlation with their spatial extent. After removing the possible radio power effect, the difference of size in low- and high-density environments persists. In the global environment analyses, the majority (86%) of high-density ERAGN reside in the cluster/group environment. In addition, ERAGN in the cluster/group central regions are preferentially compact with a small scatter in size, compared to those in the cluster/group intermediate regions and fields. In conclusion, our data appear to support the interpretation that the dense intracluster gas in the central regions of galaxy clusters plays a major role in confining the spatial extent of radio jets.
Conventional CNNs for texture synthesis consist of a sequence of (de)-convolution and up/down-sampling layers, where each layer operates locally and lacks the ability to capture the long-term structural dependency required by texture synthesis. Thus,
they often simply enlarge the input texture, rather than perform reasonable synthesis. As a compromise, many recent methods sacrifice generalizability by training and testing on the same single (or fixed set of) texture image(s), resulting in huge re-training time costs for unseen images. In this work, based on the discovery that the assembling/stitching operation in traditional texture synthesis is analogous to a transposed convolution operation, we propose a novel way of using transposed convolution operation. Specifically, we directly treat the whole encoded feature map of the input texture as transposed convolution filters and the features self-similarity map, which captures the auto-correlation information, as input to the transposed convolution. Such a design allows our framework, once trained, to be generalizable to perform synthesis of unseen textures with a single forward pass in nearly real-time. Our method achieves state-of-the-art texture synthesis quality based on various metrics. While self-similarity helps preserve the input textures regular structural patterns, our framework can also take random noise maps for irregular input textures instead of self-similarity maps as transposed convolution inputs. It allows to get more diverse results as well as generate arbitrarily large texture outputs by directly sampling large noise maps in a single pass as well.
Quasar outflows carry mass, momentum and energy into the surrounding environment, and have long been considered a potential key factor in regulating the growth of supermassive black holes and the evolution of their host galaxies. A crucial parameter
for understanding the origin of these outflows and measuring their influence on their host galaxies is the distance (R) between the outflow gas and the galaxy center. While R has been measured in a number of individual galaxies, its distribution remains unknown. Here we report the distributions of R and the kinetic luminosities of quasars outflows, using the statistical properties of broad absorption line variability in a sample of 915 quasars from the Sloan Digital Sky Surveys. The mean and standard deviation of the distribution of R are 10^{1.4+/-0.5} parsecs. The typical outflow distance in this sample is tens of parsec, which is beyond the theoretically predicted location (0.01 ~ 0.1 parsecs) where the accretion disc line-driven wind is launched, but is smaller than the scales of most outflows that are derived using the excited state absorption lines. The typical value of the mass-flow rate is of tens to a hundred solar masses per year, or several times the accretion rate. The typical kinetic-to-bolometric luminosity ratio is a few per cent, indicating that outflows are energetic enough to influence the evolution of their host galaxies.
The ability to edit materials of objects in images is desirable by many content creators. However, this is an extremely challenging task as it requires to disentangle intrinsic physical properties of an image. We propose an end-to-end network archite
cture that replicates the forward image formation process to accomplish this task. Specifically, given a single image, the network first predicts intrinsic properties, i.e. shape, illumination, and material, which are then provided to a rendering layer. This layer performs in-network image synthesis, thereby enabling the network to understand the physics behind the image formation process. The proposed rendering layer is fully differentiable, supports both diffuse and specular materials, and thus can be applicable in a variety of problem settings. We demonstrate a rich set of visually plausible material editing examples and provide an extensive comparative study.
Ultravoilet (UV) absorption lines provide abundant spectroscopic information enabling the probe of the physical conditions in AGN outflows, but the outflow radii (and the energetics consequently) can only be determined indirectly. In this paper, we p
resent the first direct test of these determinations using integral field unit (IFU) spectroscopy. We have conducted Gemini IFU mapping of the ionized gas nebulae surrounding two AGNs, whose outflow radii have been constrained by UV absorption line analyses. In Mrk 509, we find a quasi-spherical outflow with a radius of 1.2 kpc and a velocity of $sim290$ km s$^{-1}$, while IRAS F04250$-$5718 is driving a biconical outflow extending out to 2.9 kpc, with a velocity of $sim580$ km s$^{-1}$ and an opening angle of $sim70^{circ}$. The derived mass flow rate is $sim5$ and $>1$ M$_{odot}$ yr$^{-1}$, respectively, and the kinetic luminosity is $gtrsim1times10^{41}$ erg s$^{-1}$ for both. Adopting the outflow radii and geometric parameters measured from IFU, absorption line analyses would yield mass flow rates and kinetic luminosities in agreement with the above results within a factor of $sim2$. We conclude that the spatial locations, kinematics and energetics revealed by this IFU emission-line study are consistent with pre-existing UV absorption line analyses, providing a long-awaited direct confirmation of the latter as an effective approach for characterizing outflow properties.
We present HST/WFC3 narrow-band imaging of the starburst galaxy M83 targeting the hydrogen recombination lines (H$beta$, H$alpha$ and Pa$beta$), which we use to investigate the dust extinction in the HII regions. We derive extinction maps with 6 pars
ec spatial resolution from two combinations of hydrogen lines (H$alpha$/H$beta$ and H$alpha$/Pa$beta$), and show that the longer wavelengths probe larger optical depths, with $A_V$ values larger by $gtrsim$1 mag than those derived from the shorter wavelengths. This difference leads to a factor $gtrsim$2 discrepancy in the extinction-corrected H$alpha$ luminosity, a significant effect when studying extragalactic HII regions. By comparing these observations to a series of simple models, we conclude that a large diversity of absorber/emitter geometric configurations can account for the data, implying a more complex physical structure than the classical foreground dust screen assumption. However, most data points are bracketed by the foreground screen and a model where dust and emitters are uniformly mixed. When averaged over large ($gtrsim$100--200 pc) scales, the extinction becomes consistent with a dust screen, suggesting that other geometries tend to be restricted to more local scales. Moreover, the extinction in any region can be described by a combination of the foreground screen and the uniform mixture model with weights of 1/3 and 2/3 in the center ($lesssim$2 kpc), respectively, and 2/3 and 1/3 for the rest of the disk. This simple prescription significantly improves the accuracy of the dust extinction corrections and can be especially useful for pixel-based analyses of galaxies similar to M83.
