text stringlengths 29 1.99k | image imagewidth (px) 384 384 |
|---|---|
Improvements by the edge-guided inference algorithm on the six benchmark datasets, respectively | |
Improvements by the edge-guided inference algorithm on the six benchmark datasets, respectively | |
Visualization of the resulting salient maps of edges by our method, DCL <cit.>, NLDF <cit.> and DSS <cit.>, respectively. | |
Speed of different methods, including the DSS <cit.>, ELD <cit.>, DHS <cit.>, DCL <cit.>, AMU <cit.>, WSS <cit.>, C2S <cit.> and our method. | |
Illustration of our SE2Net. Firstly, we generate the low-level and high-level features through a backbone network. Secondly, the initial salient maps of edges and regions are learned from the low-level and high-level features in the first stage, respectively. Thirdly, the predicted maps of edges and regions along with ... | |
Visualization of the resulting salient maps of regions by our method, UCF <cit.>, AMU <cit.>, DSS <cit.>, BRN <cit.>, SRM <cit.> and PAGRN <cit.>, respectively. | |
GSC structure for blind beamforming | |
Analytical MSE versus simulated performance for the CCM-RLS-GSC algorithm with the TAVFF mechanism (number of users $K=5$, input $SNR = 15$dB). | |
Output SINR against the number of snapshots (number of users $K=5$, input $SNR = 15$dB). | |
Output SINR against the number of snapshots (number of users $K=5$, input $SNR = 15$dB). | |
SINR performance in a nonstationary environment (For the first stage, number of users is $K=5$, for the second stage, number of users is $K=7$, input $SNR = 15$dB). | |
Coordinate system for the calculation of the intensity distribution in the region of focus. | |
Error in $\Delta x_{\rm FWHM}$ approximated by Eq. (<ref>) with respect to the exact one by Eqs. (<ref>) and (<ref>) as a function of NA. | |
Numerical factor $\eta(\alpha)$ in Eq. (<ref>). | |
Dependence of transverse and longitudinal FWHM values, $\Delta x_{\rm FWHM}$ and $\Delta z_{\rm FWHM}$, respectively, on $w_0$ of the incident Gaussian beam. Vertical dotted lines indicate $w_0=R/2$ and $R$. (a)-(b) NA=0.4, (c)-(d) NA=0.75, and (e)-(f) NA=0.95. | |
Experimental setup for measuring the profile of the beam focused by an objective lens. L1, L2, L3: lenses, BS: beam splitter, TS1: translation stage driven by a closed-loop-feedback stepper motor, TS2: translation stage driven by closed-loop-feedback PZT actuators, OL: objective lens, CCD: charge-coupled device detecto... | |
Scanning electron microscopy image of the pinhole ($\phi=0.5 \pm 0.05 \mu$m) used as an intensity probe in our experiment. | |
Observed $xz$-profile in the focal region for NA=0.95 objective lens. The image covers a scan area of 2.5 $\mu$m $\times$ 6 $\mu$m. | |
Dependence of transverse ($x$) and longitudinal ($z$) FWHM values on $w_0$ of the incident Gaussian beam. Unconvoluted FWHM's obtained from Eq. (<ref>) are represented by solid lines whereas the convoluted FWHM's given by Eq. (<ref>) are drawn as dashed lines. Experimental results are marked by square dots with error b... | |
Left) Transmitted light optical image of U2-20 GCP. Right) Bright-field TEM image of enstatite ribbon. | |
Left) Transmitted light optical image of U2-20 GCP. Right) Bright-field TEM image of enstatite ribbon. | |
Secondary electron images of the enstatite ribbon mounted for O isotope analyses. The ribbon stuck to the Au mount nearly perpendicular to its surface. | |
Scanning ion image of $^{16}$O (counts per second). | |
Oxygen isotopic composition of enstatite ribbon (orange) with 1$\sigma$ and 2$\sigma$ uncertainties compared to the Sun <cit.>, Efremovka AOA <cit.>, Vigarano CAI <cit.>, comet Wild 2 fines <cit.>, and LL3 chondrules <cit.>. Terrestrial fractionation line and slope-one line are shown in gray. | |
Effect of learning rate on partial annotations versus full annotations on Minneapple and Bees. | |
Effect of learning rate on partial annotations versus full annotations on Minneapple and Bees. | |
uses two streams of inputs: input image and query patches of existing partial annotations. During training, positive and negative query patches are sampled from the query pool, passed through the CNN backbone, and then to the decoder after a linear transformation. The network detects objects corresponding to the class ... | |
Evaluation: Visualizations of detections when different query patches (yellow) from the same image are passed through . Left: query patch of a bird predicts the missing birds. Right: query patch of a sheep predicts the missing sheep. | |
Evaluation: Visualizations of detections when different query patches (yellow) from the same image are passed through . Left: query patch of a bird predicts the missing birds. Right: query patch of a sheep predicts the missing sheep. | |
Left: outperforms all baselines with similar annotation budget on Minneapple <cit.> dataset. PI refers to the partial images setting and PA to the partial annotations setting. Middle: an example with partial annotation (only 1 out of more than 50 instances is annotated). Right: predicts the missing annotations. | |
Left: outperforms all baselines with similar annotation budget on Minneapple <cit.> dataset. PI refers to the partial images setting and PA to the partial annotations setting. Middle: an example with partial annotation (only 1 out of more than 50 instances is annotated). Right: predicts the missing annotations. | |
Left: outperforms all baselines with similar annotation budget on Minneapple <cit.> dataset. PI refers to the partial images setting and PA to the partial annotations setting. Middle: an example with partial annotation (only 1 out of more than 50 instances is annotated). Right: predicts the missing annotations. | |
ELEGANT + CIR performance of task 1 for two images face attribute transfer | |
ELEGANT + CIR performance of task 2 for face image generation by exemplars | |
I2I-Dis + CIR performance of diverse image-to-image translation | |
GZS-Net + CIR performance of interpolation-based attribute controllable synthesis | |
Disentanglement Evaluation by Correlation Coefficient. Intra-attribute correlation increases with CIR (GZS-Net (top): $7.2\%$, ELEGANT (bottom): $3.2\%$) while inter-attribute decreases (GZS-Net: $60.9\%$, ELEGANT: $3.1\%$). | |
The influence of bias shown by Grad-Cam | |
(a-c) Intuitive understanding of Controllable Interpolation Regularization (CIR). (a) Only encourage controllable disentangled representation with general Mutual Information (MI) constrain method: maximize the MI between the same attribute across latent and observation domains while minimizing the MI between the differ... | |
(a) Directly use Controllable Interpolation in GZS-Net (b) Architecture of GZS-Net + CIR (c) Convexity optimization with Linear Interpolation (middle) and Boundary Random Interpolation. | |
Towards controllable exploration direction | |
More examples of ELEGANT+CIR (E+CIR) performance of task 1 for two images face attribute transfer | |
ELEGANT + CIR Performance of task 2 for face image generation by exemplars | |
I2I-Dis + CIR performance of diverse image-to-image translation | |
More results of GZS-Net + CIR performance of interpolation-based attribute controllable synthesis | |
Controllable mining novel background and font color by interpolation in latent space. | |
Mining new attribute values with UDV | |
Feature Cross Attention module. Feature Cross Attention module structure | |
Feature Cross Attention module. Channel attention (CA) block | |
Architecture of the Cross Attention Network. | |
Visualization results on the Cityscapes dataset. Image | |
Visualization results on the Cityscapes dataset. CANet$^{3}$ | |
Visualization results on the Cityscapes dataset. Groundtruth | |
Comparison of {existing methods} and {proposed W2PGNN} to answer when to pre-train GNNs. | |
Illustration of our proposed framework W2PGNN to answer when to pre-train GNNs. | |
{Pre-training feasibility vs. the best downstream performance on node classification when the selection buget is 2.} | |
Pre-training feasibility vs. the best downstream performance on node classification when the selection buget is 3. | |
Pre-training feasibility vs. the best downstream performance on graph classification when the selection buget is 2 and 3. | |
Pre-training feasibility vs. the best downstream performance on graph classification when the selection buget is 2 and 3. | |
Comparison between interleaving learning and block learning. In interleaving learning, we perform task 1 for a short while, then move to task 2, then task 3. Afterwards, we move from task 3 back to task 1. This process iterates where each task is performed for a short time period before switching to another task. In co... | |
How classification errors on CIFAR-100 and CIFAR-10 change as $\lambda$ increases. | |
How classification errors on CIFAR-100 and CIFAR-10 change as $\lambda$ increases. | |
How classification errors on CIFAR-100 and CIFAR-10 change as the number of interleaving rounds $M$ increases. | |
How classification errors on CIFAR-100 and CIFAR-10 change as the number of interleaving rounds $M$ increases. | |
Initial and final clouds of points | |
First neural network interface | |
Second neural network interface | |
Main block to create the mass centers and the cloud of points | |
Task 1. Unfinished function to create the cloud of points | |
Task 2. Unfinished function to calculate the distances $A$ | |
Task 3. Unfinished function to calculate the weight update | |
The success probabilities of our fermionic, bosonic and Ising quantum annealers with the total annealing time $\mathcal{T}=50$. (a) The instance-averaged success probabilities of the three quantum annealers for the problem sizes $N = 4\times 3, 4\times 4, 4\times 5$. The individual success probabilities for the problem... | |
(a) The solution degeneracy distribution of our 1000 problem instances for the system size $N=4\times 4$. (b) The success probabilities of our fermionic, bosonic and Ising quantum annealers for the problem size $N=4\times 4$, which are averaged over the instances with the same solution degeneracy $D$ and presented as f... | |
(a) The relevant gaps of the interpolating Hamiltonians followed by the fermionic, the bosonic and the Ising quantum annealer to solve the problem with size $N=4\times 4$, which are averaged over the instances with the same solution degeneracy $D$ and presented as functions of $D$. The lowest twelve energy levels of th... | |
The ground-state fidelity susceptibilities per site along the annealing process of the fermionic, the bosonic and the Ising quantum annealers, which are averaged over the instances with $N=4\times 4$ and $D=2$. | |
(a) The ground-state glass order and (b) the average glass order of the lowest twelve eigenstates along the annealing process of the fermionic, the bosonic and the Ising quantum annealers. Both are averaged over the instances with $N=4\times 4$ and $D=2$. | |
(a) The effective dimensions of the time-evolving state $|\psi(t)\rangle$ (solid lines), with the ones of the instantaneous ground state $|\psi_{\text{g}}(t)\rangle$ (dotted lines) as an adiabatic reference, and (b) the instantaneous values of the success probability (solid lines) and the ground state probability (dott... | |
Design of investigated {pmsm} containing three polepairs and $36$ slots. | |
Visualization of GPR-Hybrid approach with accepted (blue), not accepted (red) and critical (orange) sample points. | |
Visualization of GPR-Hybrid approach with accepted (blue), not accepted (red) and critical (orange) sample points. | |
Visualization of the pareto-front for a bi-objective maximization problem. | |
Geometrical depiction of uncertain rotor design parameters based on <cit.>. | |
Geometrical depiction of deterministic motor design parameters $\mathbf{x}$. | |
Comparison of FE model evaluations and {gpr} standard deviation $\sigma_{\text{GPR}}$ propagation for the two proposed test cases. Iteration $i$ refers to th $i$-th {mc} sample point within the {gpr}-Hybrid approach. | |
Propagation of the yield estimate throughout the optimization for the $\varepsilon$-constraint method. | |
Propagation of the PM surface dimensions throughout the optimization for the $\varepsilon$-constraint method. | |
Propagation of the yield estimate throughout the optimization for the weighted sum method. | |
Propagation of the {pm} surface dimensions throughout the optimization for the weighted sum method. | |
Demonstration of exploration phase where the best performing starting point is highlighted in blue. | |
Comparison of the propagation of the yield estimate throughout the optimization for the weighted sum method with and without the multi-start procedure. | |
Optimized design (multi-start method). Here, the fille rectangle displays the optimized design and the contours the original design dimensions. | |
Visualization of the pareto front of the 8th generation of the genetic algorithm for the stated problem similar to Figure <ref>. NSGA-II algorithm converges to single solution marked with red square. | |
The AdS description of the three-point function. The thick curve represents a geodesic of the D-brane dual to the determinant operators while the wavy line denotes a closed string dual to the single trace operator. On the worldsheet, it corresponds to an overlap with a boundary state. | |
The partition function $Z(J,R)$ evaluated in two different channels. The left/right panel denote the closed-/open- string channel of the same partition function. To compute the overlap $\langle \mathcal{G}|\Omega\rangle$, we take the limit where $J$ is finite and $R\to\infty$. | |
The boundary Yang-Baxter equation. | |
Watson's equation for the two-particle overlap. | |
Decoupling equation for the two-particle overlap. | |
The bABA and its unfolding interpretation. The structure of the reflection matrix allows us to map the bABA to an ABA of a closed chain with a single $\mathfrak{psu}(2|2)$ symmetry. | |
Expected planet yield from a SPIRou RV planet search over 300 nights and including 360 mid-to-late M dwarfs. Filled circled indicate detected planets, open circles undetected ones and red circles (both filled and open) represent transiting planets. A rough indication of the limits of the habitable zone, both in mass an... | |
Schematic view of the SPIRou instrument main components. The Cassegrain unit contains polarimetric analyzers injecting star light into two 35 m-long fibers. The main spectrograph, to be installed at the CFHT coudé room, consists of a 3m-long cryostat (see also figure <ref>). A calibration unit can inject light form a ... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.