HDR-Net-Fusion: Real-time 3D dynamic scene reconstruction with a hierarchical deep reinforcement network

Reconstructing dynamic scenes with commodity depth cameras has many applications in computer graphics, computer vision, and robotics. However, due to the presence of noise and erroneous observations from data capturing devices and the inherently ill-posed nature of non-rigid registration with insufficient information, traditional approaches often produce low-quality geometry with holes, bumps, and misalignments. We propose a novel 3D dynamic reconstruction system, named HDR-Net-Fusion, which learns to simultaneously reconstruct and refine the geometry on the fly with a sparse embedded deformation graph of surfels, using a hierarchical deep reinforcement (HDR) network. The latter comprises two parts: a global HDR-Net which rapidly detects local regions with large geometric errors, and a local HDR-Net serving as a local patch refinement operator to promptly complete and enhance such regions. Training the global HDR-Net is formulated as a novel reinforcement learning problem to implicitly learn the region selection strategy with the goal of improving the overall reconstruction quality. The applicability and efficiency of our approach are demonstrated using a large-scale dynamic reconstruction dataset. Our method can reconstruct geometry with higher quality than traditional methods.

Improved dynamic imaging of multiphase flow by constrained tomographic reconstruction

Dynamic tomography has become an important technique to study fluid flow processes in porous media. The use of laboratory X-ray tomography instruments is, however, limited by their low X-ray brilliance. The prolonged exposure times, in turn, greatly limit temporal resolution. We have developed a tomographic reconstruction algorithm that maintains high image quality, despite reducing the exposure time and the number of projections significantly. Our approach, based on the Simultaneous Iterative Reconstruction Technique, mitigates the problem of few and noisy exposures by utilising a high-quality scan of the system before the dynamic process is started. We use the high-quality scan to initialise the first time step of the dynamic reconstruction. We further constrain regions of the dynamic reconstruction with a segmentation of the static system. We test the performance of the algorithm by reconstructing the dynamics of fluid separation in a multiphase system. The algorithm is compared quantitatively and qualitatively with several other reconstruction algorithms and we show that it can maintain high image quality using only a fraction of the normally required number of projections and with a substantially larger noise level. By robustly allowing fewer projections and shorter exposure, our algorithm enables the study of faster flow processes using laboratory tomography instrumentation but it can also be used to improve the reconstruction quality of dynamic synchrotron experiments.

Phase reconfiguration of multivalent nickel sulfides in hydrogen evolution

Deciphering the dynamic evolution of catalysts’ atomic and electronic structure in operating conditions is pivotal for unraveling the activity origin and improving catalyst design. Earth-abundant transition metal catalysts have shown efficient catalytic efficiency and are attractive due to sustainable and economic considerations. However, the dynamic evolution process during their whole service time remains elusive, which is greatly complicated by the multiple component and valence states as well as the structural complexity of materials. Here in this work, we investigated the atomic-scale evolution of multivalent nickel-based sulfides (from NiS2 to α-NiS, β-NiS and Ni3S4) as model catalysts for hydrogen evolution reaction (HER), via operando Raman and X-ray absorption spectroscopies corroborated by theoretical calculations. Dynamic reconstruction propagating from surface to bulk, mediated by sulphur vacancy, has been demonstrated for these materials, all with the terminated Ni3S2 phase on catalyst surface being responsible for subsequent catalysis. Partial Fe substitution prompts such reconfiguration process and hence improves HER performance, which establishes the dynamic working mechanism of widely-adopted doping strategy. We unprecedentedly reveal the dynamic reconstruction with lower valence state tendency of transition metals in the catalytically terminated phase during HER, and the life-time dynamic correlation between structure and activity, providing insights into future catalyst design.