Search

In this paper, we tackle one-shot texture retrieval: given an example of a new reference texture, detect and segment all the pixels of the same texture category within an arbitrary image. To address this problem, we present an OS-TR network to encode both reference and query image, leading to achieve texture segmentation towards the reference category. Unlike the existing texture encoding methods that integrate CNN with orderless pooling, we propose a directionality-aware module to capture the texture variations at each direction, resulting in spatially invariant representation. To segment new categories given only few examples, we incorporate a self-gating mechanism into relation network to exploit global context information for adjusting per-channel modulation weights of local relation features. Extensive experiments on benchmark texture datasets and real scenarios demonstrate the above-par segmentation performance and robust generalization across domains of our proposed method., Comment: ijcai2019-lastest

The Feynman amplitude for the decay of the $J/\psi$ meson into baryon-antibaryon can be written as a sum of three sub-amplitudes: a purely strong, a purely electromagnetic and a mixed strong-electromagnetic. Assuming that the strong and mixed strong-electromagnetic sub-amplitudes have the same phase, the branching ratio of the decay contains an interference term that depends on the relative phase $\varphi$ between strong and electromagnetic sub-amplitudes. In this work we calculate this phase, by using an effective strong Lagrangian density and considering, as final states, pairs of baryons, $\mathcal{B}\overline{\mathcal{B}}$, belonging to the spin-1/2 SU(3) octet. Moreover, we obtain the purely strong, purely electromagnetic and mixed strong-electromagnetic contributions to the total branching ratio and hence the moduli of the corresponding sub-amplitudes, for each pair of baryons. Of particular interest is the mixed strong-electromagnetic contribution that, not only is determined for the first time, but it is proven to be crucial, in the framework of our model, for the correct description of the decay mechanism. Finally we use the purely electromagnetic branching ratio to calculate the Born non-resonant cross section of the annihilation processes $e^+ e^- \to \mathcal{B}\overline{\mathcal{B}}$ at the $J/\psi$ mass. By taking advantage from all available data, we obtain the relative phase between strong and electromagnetic sub-amplitudes: $\varphi = (73\pm 8)^\circ$., Comment: 8 pages, 1 figure

We derive the mass weighted total density slopes within the effective (half-light) radius, $\gamma'$, for more than 2000 nearby galaxies from the SDSS-IV MaNGA survey using Jeans-anisotropic-models applied to IFU observations. Our galaxies span a wide range of the stellar mass ($10^9$ $M_{\rm \odot}< M_* < 10^{12}$ M$_{\odot}$) and the velocity dispersion (30 km/s $< \sigma_v <$ 300 km/s). We find that for galaxies with velocity dispersion $\sigma_v>100$ km/s, the density slope has a mean value $\langle \gamma^{\prime} \rangle = 2.24$ and a dispersion $\sigma_{\gamma}=0.22$, almost independent of velocity dispersion. A clear turn over in the $\gamma'-\sigma_v$ relation is present at $\sigma\sim 100$ km/s, below which the density slope decreases rapidly with $\sigma_v$. Our analysis shows that a large fraction of dwarf galaxies (below $M_* = 10^{10}$ M$_{\odot}$) have total density slopes shallower than 1, which implies that they may reside in cold dark matter halos with shallow density slopes. We compare our results with that of galaxies in hydrodynamical simulations of EAGLE, Illustris and IllustrisTNG projects, and find all simulations predict shallower density slopes for massive galaxies with high $\sigma_v$. Finally, we explore the dependence of $\gamma'$ on the positions of galaxies in halos, namely centrals vs. satellites, and find that for the same velocity dispersion, the amplitude of $\gamma'$ is higher for satellite galaxies by about 0.1., Comment: 11 pages, 10 figures, submitted to MNRAS

We study the track reconstruction algorithms of the CEPC luminometer. Depend on the current geometry design, the conventional track reconstruction method is applied, but it suffers the energy leakage problem when tracks falling into the tile gaps regions. To solve this problem, a novel reconstruction method based on deep neural networks has been investigated, and the reconstruction efficiency has been improved significantly, as well as the energy and direction resolutions. This new reconstruction method is proposed to replace the conventional one for the CEPC luminometer., Comment: 8 pages, 12 figures, 2 tables

In recent years, there is a surge on machine learning applications in industry. Many of them are based on popular AI frameworks like Tensorflow, Torch, Caffe, or MxNet, etc, and are enpowered by accelerator platforms such as GPUs. One important challenge of running Tensorflow computations on GPUs is the fine granularity problem, namely, FLOPS of individual ops are far from enough to fully exploit the computing power of underlying accelerators. The XLA framework provides a solid foundation to explore this problem further. In this paper, we propose FusionStitching, a novel, comprehensive Op fusion and code generation system to stitch computations into large GPU kernels. Experimental results on four public models and two of our large inhouse applications show another 55% (geometric mean) reduction of GPU kernel launches, compared to the XLA fusion baseline. This increases the E2E performance of both of our latency critical inhouse applications up to 20%., Comment: 11 pages, 8 figures