Search

Transformer-based Super-Resolution (SR) models have recently advanced image reconstruction quality, yet challenges remain due to computational complexity and an over-reliance on large patch sizes, which constrain fine-grained detail enhancement. In this work, we propose TaylorIR to address these limitations by utilizing a patch size of 1x1, enabling pixel-level processing in any transformer-based SR model. To address the significant computational demands under the traditional self-attention mechanism, we employ the TaylorShift attention mechanism, a memory-efficient alternative based on Taylor series expansion, achieving full token-to-token interactions with linear complexity. Experimental results demonstrate that our approach achieves new state-of-the-art SR performance while reducing memory consumption by up to 60% compared to traditional self-attention-based transformers.

We introduce a LiDAR inertial odometry (LIO) framework, called LiPO, that enables direct comparisons of different iterative closest point (ICP) point cloud registration methods. The two common ICP methods we compare are point-to-point (P2P) and point-to-feature (P2F). In our experience, within the context of LIO, P2F-ICP results in less drift and improved mapping accuracy when robots move aggressively through challenging environments when compared to P2P-ICP. However, P2F-ICP methods require more hand-tuned hyper-parameters that make P2F-ICP less general across all environments and motions. In real-world field robotics applications where robots are used across different environments, more general P2P-ICP methods may be preferred despite increased drift. In this paper, we seek to better quantify the trade-off between P2P-ICP and P2F-ICP to help inform when each method should be used. To explore this trade-off, we use LiPO to directly compare ICP methods and test on relevant benchmark datasets as well as on our custom unpiloted ground vehicle (UGV). We find that overall, P2F-ICP has reduced drift and improved mapping accuracy, but, P2P-ICP is more consistent across all environments and motions with minimal drift increase., Comment: Submitted to ICRA 2025

Magnetic fields are the primary driver of the plasma thermodynamics in the upper solar atmosphere, especially in the corona. However, magnetic field measurements in the solar corona are sporadic, thereby limiting us from the complete understanding of physical processes occurring in the coronal plasma. In this paper, we explore the diagnostic potential of a coronal emission line in the extreme-ultraviolet (EUV), i.e., Ne VIII 770 \AA to probe the coronal magnetic fields. We utilize 3D 'Magneto-hydrodynamic Algorithm outside a Sphere' (MAS) models as input to the FORWARD code to model polarization in Ne VIII line produced due to resonance scattering, and interpret its modification due to collisions and the magnetic fields through the Hanle effect. The polarization maps are synthesized both on the disk as well as off-the-limb. The variation of this polarization signal through the different phases of solar cycle 24 and the beginning phase of solar cycle 25 is studied in order to understand the magnetic diagnostic properties of this line owing to different physical conditions in the solar atmosphere. The detectability of the linear polarization signatures of the Hanle effect significantly improves with increasing solar activity, consistently with the increase in the magnetic field strength and the intensity of the mean solar brightness at these wavelengths. We finally discuss the signal-to-noise ratio (SNR) requirements by considering realistic instrument designs., Comment: 16 pages, 11 figures. Accepted for publication in the Astrophysical Journal

We present spectropolarimetric observations of an active region recorded simultaneously in the H$\alpha$ Ca II 8662 {\AA} lines. The sunspot exhibits multiple structures, including a lightbridge and a region where Ca II 8662 {\AA} line core is in emission. Correspondingly, the H$\alpha$ line core image displays brightening in the emission region, with the spectral profiles showing elevated line cores. The stratification of the line-of-sight magnetic field is inferred through non-LTE multiline inversions of the Ca II 8662 {\AA} line and the weak field approximation over the H$\alpha$ line. The field strength inferred from the H$\alpha$ line core is consistently smaller than that inferred from inversions at $\log \tau_{500}$ = $-$4.5. However, the study finds no correlation between the WFA over the core of the H$\alpha$ line and that inferred from inversions at $\log \tau_{500}$ = $-$4.5. In regions exhibiting emission features, the morphology of the magnetic field at $\log \tau_{500}$ = $-$4.5 resembles that at $\log \tau_{500}$ = $-$1, with slightly higher or comparable field strengths. The magnetic field morphology inferred from the core of the H$\alpha$ line is also similar to that inferred from the full spectral range of the H$\alpha$ line in the emission region. The field strength inferred in the lightbridge at $\log \tau_{500}$ = $-$1 is smaller than the surrounding umbral regions and comparable at $\log \tau_{500}$ = $-$4.5. Similarly, the field strength inferred in the lightbridge from the WFA over the H$\alpha$ line appears lower compared to the surrounding umbral regions., Comment: accepted to be published in ApJ on 4th June, 2024, 14 pages, 7 figures

Software reliability growth models (SRGM) enable failure data collected during testing. Specifically, nonhomogeneous Poisson process (NHPP) SRGM are the most commonly employed models. While software reliability growth models are important, efficient modeling of complex software systems increases the complexity of models. Increased model complexity presents a challenge in identifying robust and computationally efficient algorithms to identify model parameters and reduces the generalizability of the models. Existing studies on traditional software reliability growth models suggest that NHPP models characterize defect data as a smooth continuous curve and fail to capture changes in the defect discovery process. Therefore, the model fits well under ideal conditions, but it is not adaptable and will only fit appropriately shaped data. Neural networks and other machine learning methods have been applied to greater effect [5], however limited due to lack of large samples of defect data especially at earlier stages of testing.