High-resolution Reconstruction Research Articles

Microplastics deposition in Arctic sediments of Greenland increases significantly after 1950

Marine sediments are archives of environmental change with pollutants potentially acting as chronographic markers of the Anthropocene. Particularly, the vertical transport and burial, as well as slow degradation rate of microplastics, indicate an eventual incorporation into the geological record. A high-resolution reconstruction of microplastics records requires high sedimentation rates, and in Disko Bay (Greenland), this quality coincides with a need for plastic pollution data. Here, a Greenlandic sediment core dated back to 1930 ± 2 was reconstructed for microplastics accumulation via micro-Fourier transform infrared spectroscopy. We show 85 years of fluctuating microplastics accumulation (987 − 16,645 particles Kg−1) down to 20 µm and diversified into eight polymers. Polyethylene (47%) and polypropylene (32%) were more abundantly present through time. Microplastics accumulation increases significantly after 1950 along with major socio-economic development in the area, suggesting an influence from regional stressors. Regional microplastics reconstructions must therefore be considered in the pursuit of an Anthropocene global plastic horizon.

High-resolution ocean color reconstruction and analysis focusing on Kd490 via machine learning model integration of MODIS and Sentinel-2 (MSI)

High-resolution ocean color reconstruction and analysis focusing on Kd490 via machine learning model integration of MODIS and Sentinel-2 (MSI)

An adaptive parameter decoupling algorithm-based image reconstruction model (ADAIR) for rapid golden-angle radial DCE-MRI.

The acceleration of magnetic resonance imaging (MRI) acquisition is crucial for both clinical and research applications, particularly in dynamic MRI. Existing compressed sensing methods, despite being effective for fast imaging, face limitations such as the need for incoherent sampling and residual noise, which restrict their practical use for rapid MRI. To overcome these challenges, we propose a novel image reconstruction framework that integrates the MRI physical model with a flexible, self-adjusting, decoupling data-driven model. We validated this method through experiments using both simulated and in vivo dynamic contrast-enhanced MRI datasets. The experimental results demonstrate that the proposed framework achieves high spatial and temporal resolution reconstructions. Additionally, when compared to state-of-the-art image reconstruction approaches, our method significantly enhances acceleration capabilities, enabling sparse and rapid imaging with high resolution. Our proposed framework offers a promising solution for real-time imaging and image-guided radiation therapy applications by providing superior performance in achieving high spatial and temporal resolution reconstructions, thus addressing the limitations of existing compressed sensing schemes.

20 µm resolution multipixel ghost imaging with high-energy x-rays

Hard x-ray imaging is indispensable across diverse fields owing to its high penetrability. However, the resolution of traditional x-ray imaging modalities, such as computed tomography (CT) systems, is constrained by factors including beam properties, the limitations of optical components, and detection resolution. As a result, the typical resolution in commercial imaging systems that provide full-field imaging is limited to a few hundred microns, and scanning CT systems are too slow for many applications. This study advances high-photon-energy imaging by extending the concept of computational ghost imaging to multipixel ghost imaging with x-rays. We demonstrate a remarkable resolution of approximately 20 µm for an image spanning 0.9 by 1 cm2, comprised of 400,000 pixels and involving only 1000 realizations. Furthermore, we present a high-resolution CT reconstruction using our method, revealing enhanced visibility and resolution. Our achievement is facilitated by an innovative x-ray lithography technique and the computed tiling of images captured by each detector pixel. Importantly, this method maintains reasonable timeframes and can be scaled up for larger images without sacrificing the short measurement time, thereby opening intriguing possibilities for noninvasive high-resolution imaging of small features that are invisible with the present modalities.

High-accuracy direction measurement and high-resolution computational spectral reconstruction based on photonic crystal array

Portable and wearable miniaturized spectrometers play a crucial role in various fields. In this paper, we present a method for simultaneously realizing high-accuracy direction measurement and high-resolution computational spectral reconstruction based on the angle sensitivity of conventional photonic crystals (PCs), wherein an optical filter array is composed of multiple one-dimensional PCs. The high-angle sensitivity of PCs results in angle-dependent optical spectra. When these spectra with different angles are used to reconstruct the target spectra in an unknown direction and the interval between adjacent angles is sufficiently small, the accurate direction of the target can be automatically identified. Moreover, the computational spectra still have high resolution over a wide range of incidences. The computational spectra under arbitrary polarizations can also be recognized based on the polarization dependence of the PCs at an oblique incidence. Our research results are significant for engineering a new miniaturized comprehensive computational spectrometer with target-direction perception and omnidirectional spectral reconstruction abilities.

Substructures within Substructures in the Complex Postmerging System A514 Unveiled by High-resolution Magellan/Megacam Weak Lensing

Abell 514 (A514) at z = 0.071 is an intriguing merging system exhibiting highly elongated (∼1 Mpc) X-ray features and three large-scale (300 ∼ 500 kpc) bent radio jets. To dissect this system with its multiwavelength data, it is critical to robustly identify and quantify its dark matter substructures. We present a weak-lensing analysis of A514 using deep Magellan/Megacam observations. Combining two optical band filter imaging data obtained under optimal seeing (∼0.″6) and leveraging the proximity of A514, we achieve a high source density of ∼46arcmin−2 or ∼6940 Mpc−2, which enables high-resolution mass reconstruction. We unveil the complex dark matter substructures of A514, which are characterized by the NW and SE subclusters separated by ∼0.7 Mpc, each exhibiting a bimodal mass distribution. The total mass of the NW subcluster is estimated to be M200cNW=1.08−0.22+0.25×1014M⊙ and is further resolved into the eastern ( M200cNWE=2.6−1.1+1.4×1013M⊙) and western ( M200cNWW=7.1−2.0+2.3×1013M⊙ ) components. The mass of the SE subcluster is M200cSE=1.55−0.26+0.28×1014M⊙ , which is also further resolved into the northern ( M200cSEN=2.9−1.3+1.8×1013M⊙ ) and southern ( M200cSES=8.5−2.6+3.1×1013M⊙ ) components. These four substructures coincide with the A514 brightest galaxies and are detected with significances ranging from 3.3σ to 4.7σ. Comparison of the dark matter substructures with the X-ray distribution suggests that A514 might have experienced an off-axis collision, and the NW and SE subclusters are currently near their apocenters.

Investigating the Use of Traveltime and Reflection Tomography for Deep Learning-Based Sound-Speed Estimation in Ultrasound Computed Tomography.

Ultrasound computed tomography (USCT) quantifies acoustic tissue properties such as the speed-of-sound (SOS). Although full-waveform inversion (FWI) is an effective method for accurate SOS reconstruction, it can be computationally challenging for large-scale problems. Deep learning-based image-to-image learned reconstruction (IILR) methods can offer computationally efficient alternatives. This study investigates the impact of the chosen input modalities on IILR methods for high-resolution SOS reconstruction in USCT. The selected modalities are traveltime tomography (TT) and reflection tomography (RT), which produce a low-resolution SOS map and a reflectivity map, respectively. These modalities have been chosen for their lower computational cost relative to FWI and their capacity to provide complementary information: TT offers a direct SOS measure, while RT reveals tissue boundary information. Systematic analyses were facilitated by employing a virtual USCT imaging system with anatomically realistic numerical breast phantoms. Within this testbed, a supervised convolutional neural network (CNN) was trained to map dual-channel (TT and RT images) to a high-resolution SOS map. Single-input CNNs were trained separately using inputs from each modality alone (TT or RT) for comparison. The accuracy of the methods was systematically assessed using normalized root mean squared error (NRMSE), structural similarity index measure (SSIM), and peak signal-to-noise ratio (PSNR). For tumor detection performance, receiver operating characteristic analysis was employed. The dual-channel IILR method was also tested on clinical human breast data. Ensemble average of the NRMSE, SSIM, and PSNR evaluated on this clinical dataset were 0.2355, 0.8845, and 28.33 dB, respectively.

A novel biomechanical model of the mouse forelimb predicts muscle activity in optimal control simulations of reaching movements.

Mice are key model organisms in genetics, neuroscience and motor systems physiology. Fine motor control tasks performed by mice have become widely used in assaying neural and biophysical motor system mechanisms, including lever or joystick manipulation, and reach-to-grasp tasks (Becker et al., 2019; Bollu et al., 2019; Conner at al., 2021). Although fine motor tasks provide useful insights into behaviors which require complex multi-joint motor control, there is no previously developed physiological biomechanical model of the adult mouse forelimb available for estimating kinematics (including joint angles, joint velocities, fiber lengths and fiber velocities) nor muscle activity or kinetics (including forces and moments) during these behaviors. Here we have developed a musculoskeletal model based on high-resolution imaging and reconstruction of the mouse forelimb that includes muscles spanning the neck, trunk, shoulder, and limbs using anatomical data. Physics-based optimal control simulations of the forelimb model were used to estimate in vivo muscle activity present when constrained to the tracked kinematics during mouse reaching movements. The activity of a subset of muscles was recorded via electromyography and used as the ground truth to assess the accuracy of the muscle patterning in simulation. We found that the synthesized muscle patterning in the forelimb model had a strong resemblance to empirical muscle patterning, suggesting that our model has utility in providing a realistic set of estimated muscle excitations over time when provided with a kinematic template. The strength of the resemblance between empirical muscle activity and optimal control predictions increases as mice performance improves throughout learning of the reaching task. Our computational tools are available as open-source in the OpenSim physics and modeling platform (Seth et al., 2018). Our model can enhance research into limb control across broad research topics and can inform analyses of motor learning, muscle synergies, neural patterning, and behavioral research that would otherwise be inaccessible.

Historic ocean acidification of Loch Sween revealed by correlative geochemical imaging and high-resolution boron isotope analysis of Boreolithothamniom cf. soriferum

Ocean Acidification (OA) arises from the increase in atmospheric carbon dioxide concentration following the industrial revolution. The ecological and socio-economic consequences of OA were first identified around 10–15 years ago but remain poorly understood. This is particularly true in coastal regions where local processes can have dramatic consequences on pH trends through time, obscuring and compounding the long-term effects from rising atmospheric CO2. Here we explore the possibility of generating long records of coastal ocean pH using the skeletons of widely distributed coralline algae (CA). The skeletons of these slow growing (<1 mm/year) taxa often contain micron-scale heterogeneities, making sampling for high-resolution climate reconstructions using bulk sampling techniques difficult. Here we use laser ablation coupled to inductively coupled plasma mass spectrometers to generate high-resolution 2D images of the element/calcium ratios and boron isotope composition (δ11B) of a sample of Boreolithothamniom cf. soriferum from Loch Sween in Scotland, UK where we have been monitoring temperature since 2004 and pH during 2014. By carefully correlating the geochemical images with a scanning electron microscopy image we can segment them to remove the marginal portions of the skeleton, isolating the central growth axis to generate an age model and growth rate. The δ11B-pH is significantly elevated above the seawater pH in Loch Sween (8.4 to 8.9 vs. 7.9 to 8.1) consistent with other CA that show internal pH elevation. On a seasonal scale, internal pH is negatively correlated with temperature and also exhibits a long-term decline. By removing this temperature effect, internal pH can be correlated to seawater pH during the 2014 monitoring period allowing us to reconstruct a seawater acidification trend from 2004 to 2018 of -0.018 pH units per year, 10x higher than open ocean trends but consistent with contemporaneous monitoring efforts of UK coastal waters. Reconstructed aqueous CO2 suggests that prior to ∼2008 Loch Sween was a sink of CO2 but after this date, particularly during the early summer, it was a substantial CO2 source. Comparison of reconstructed aqueous CO2 with a record of calcification rate of our sample of Boreolithothamniom cf. soriferum suggests this acidification and associated rise in local seawater pCO2 may have freed this sample from carbon limitation leading to a recent increase in calcification.

Dual-modality ghost diffraction in a complex disordered environment using untrained neural networks

We report a dual-modality ghost diffraction (GD) system to simultaneously enable high-fidelity data transmission and high-resolution object reconstruction through complex disordered media using an untrained neural network (UNN) with only one set of realizations. The pixels of a 2D image to be transmitted are sequentially encoded into a series of random amplitude-only patterns using a UNN without labels and datasets. The series of random patterns generated is sequentially displayed to interact with an object placed in a designed optical system through complex disordered media. The realizations recorded at the receiving end are used to retrieve the transmitted data and reconstruct the object at the same time. The experimental results demonstrate that the proposed dual-modality GD system can robustly enable high-fidelity data transmission and high-resolution object reconstruction in a complex disordered environment. This could be a promising step toward the development of AI-driven compact optical systems with multiple modalities through complex disordered media.

Information sparsity guided transformer for multi-modal medical image super-resolution

Multi-modal medical image super-resolution (SR) plays a vital role in enhancing the resolution of medical images, providing more detailed visuals that aid in accurate clinical diagnosis. Recently, Transformer-based super-resolution methods have significantly promoted performance improvement in this field due to their capacity to capture global dependencies. They usually process all non-overlapping patches as tokens and densely sample these tokens without screening for calculating attention mechanisms. However, this strategy ignores the spatial sparsity of medical images, resulting in redundant or even detrimental computations on less informative regions. Hence, this paper proposes a novel sparsity-guided medical image SR network, namely SG-SRNet, by exploiting the spatial sparsity characteristics of the medical images. SG-SRNet mainly consists of two components: a sparsity mask (SM) generator for image sparsity estimation, and a sparsity-guided Transformer (SGTrans) for high-resolution image reconstruction. Specially, the SM generator generates a sparsity mask by minimizing our cross-sparsity loss which can respond to the informative positions. SGTrans first screens out the informative patches according to the sparsity mask. Then, SGTrans utilizes the designed cluster-based attention to only calculate attention between information-related tokens. We perform comprehensive experiments on three datasets to show that SG-SRNet brings significant performance enhancements with low computational complexity.

Strandline records of high frequency, low magnitude drops in water level, glacial Lake Agassiz basin, Central Polk County, MN, USA

Glacial Lake Agassiz strandlines expressed in LiDAR-based digital elevation models are detailed records of past water level change that reflect outlet incision and switching. Detailed 1:24,000 scale surficial geology mapping of a small section of coastline east of Crookston, MN, reveals strandlines of the Lockhart, Emerson, and Nipigon phases. Other than the Campbell and youngest Tintah Beach, the strandlines are not identified using the naming convention established near the southern outlet (Herman, Norcross, Upham, Tintah) because of the increasing number of strandlines northwards. The strandlines, for the most part, naturally exist in elevation clusters, and are assigned to 11 strandline groups (SG). SGs 1–9 are of Lockhart age at elevations higher than the Tintah Campbell Gap (TCGap). Elevation gaps between SGs are ~1–3 m, greater than the 0.5–2 m gap between successive strandlines within a SG. The many (n = 70) small, low-relief strandlines are interpreted to record nearly continuous incision of the southern outlet sill during Lockhart Phase. Beach ridges are estimated to have formed every ~20 years. The TCGap is ~4 km wide in the study area. The Campbell Beach of Emerson Phase age consists of several embayments, beach ridges and the large Melvin Spit. Based on morphological comparison with the modern Gull Point Spit in the Lake Erie basin, the Melvin Spit took ~500 years to form, during which water level dropped 8 m. The Nipigon Phase strandlines are small discontinuous beach ridges. North-south elevation plots of the most continuous strandline in each SG records southward slopes of 1.1 to 0.19 m/km explained by differential glacioisostatic adjustment active during strandline development that decreases through time at lower elevations. The three highest elevation OSL dates agree with published ages for the older strandlines (~14.4 ± 1.5 ka), but younger dated Tintah, Campbell, and sub-Campbell beaches are 2–5 ka older than published ages. Ground penetrating radar (GPR) datasets acquired at 11 sites at frequencies between 100 and 500 MHz, representing a cumulative distance of 11 km, form the basis for a simple beach ridge formation model augmented with hand auger holes and shallow hand-dug pits. Six radar facies (RF) were identified, dominated by RF2 which consists of lakeward-dipping reflections with a downlapping lower boundary recording a depositional regressive system, and RF4 which consists of landward-dipping reflections with a downlapping lower boundary recording overwash deposits from proposed storm events. Spit orientations and GPR data indicate littoral drift was southwards with offshore bars possibly nucleating successive beach ridges upon ~1-m drops in water level. Detailed analysis of strandlines from paleo lakes offer promise for high resolution reconstructions of past water level history.

Snapshot coherent diffraction imaging across ultra-broadband spectra

Ultrafast imaging simultaneously pursuing high temporal and spatial resolution is a key technique to study the dynamics in the microscopic world. However, the broadband spectra of ultra-short pulses bring a major challenge to traditional coherent diffraction imaging (CDI), as they result in an indistinct diffraction pattern, thereby complicating image reconstruction. To address this, we introduce, to our knowledge, a new ultra-broadband coherent imaging method, and empirically demonstrate its efficacy in facilitating high-resolution and rapid image reconstruction of achromatic objects. The existing full bandwidth limitation for snapshot CDI is enhanced to ∼60% experimentally, restricted solely by our laser bandwidth. Simulations indicate the applicability of our method for CDI operations with a bandwidth as high as ∼140%, potentially supporting ultrafast imaging with temporal resolution into ∼50-attosecond scale. Even deployed with a comb-like harmonic spectrum encompassing multiple octaves, our method remains effective. Furthermore, we establish the capability of our approach in reconstructing a super-broadband spectrum for CDI applications with high fidelity. Given these advancements, we anticipate that our method will contribute significantly to attosecond imaging, thereby advancing cutting-edge applications in material science, quantum physics, and biological research.

High-Resolution Reconstruction of a C. elegans Ribosome Sheds Light on Evolutionary Dynamics and Tissue Specificity.

Caenorhabditis elegans is an important model organism for human health and disease, with foundational contributions to the understanding of gene expression and tissue patterning in animals. An invaluable tool in modern gene expression research is the presence of a high-resolution ribosome structure, though no such structure exists for C. elegans. Here we present a high-resolution single-particle cryogenic electron microscopy (cryoEM) reconstruction and molecular model of a C. elegans ribosome, revealing a significantly streamlined animal ribosome. Many facets of ribosome structure are conserved in C. elegans, including overall ribosomal architecture and the mechanism of cycloheximide, while other facets such as expansion segments and eL28 are rapidly evolving. We identify uL5 and uL23 as two instances of tissue-specific ribosomal protein paralog expression conserved in Caenorhabditis, suggesting that C. elegans ribosomes vary across tissues. The C. elegans ribosome structure will provide a basis for future structural, biochemical, and genetic studies of translation in this important animal system.

Deep learning for the super resolution of Mediterranean sea surface temperature fields

Abstract. Sea surface temperature (SST) is one of the essential variables of the Earth's climate system. Being at the air–sea interface, SST modulates heat fluxes in and out of the ocean, provides insight into several upper and interior ocean dynamical processes, and is a fundamental indicator of climate variability potentially impacting the health of marine ecosystems. Its accurate estimation and regular monitoring from space is therefore crucial. However, even if satellite infrared/microwave measurements provide much better coverage than what is achievable from in situ platforms, they cannot sense the sea surface under cloudy and rainy conditions. Large gaps are present even in merged multi-sensor satellite products, and different statistical strategies, mostly based on optimal interpolation (OI) algorithms, have thus been proposed to obtain gap-free (L4) images. These techniques, however, filter out the signals below the space–time decorrelation scales considered, significantly smoothing most of the small mesoscale and submesoscale features. Here, deep learning models, originally designed for single-image super resolution (SR), are applied to enhance the effective resolution of SST products and the accuracy of SST gradients. SR schemes include a set of computer vision techniques leveraging convolutional neural networks to retrieve high-resolution data from low-resolution images. A dilated convolutional multi-scale learning network, which includes an adaptive residual strategy and implements a channel attention mechanism, is used to reconstruct features in SST data at 1/100° spatial resolution starting from 1/16° data over the Mediterranean Sea. The application of this technique shows an improvement in the high-resolution reconstruction, capturing small-scale features and providing a root-mean-squared-difference improvement of 0.02 °C with respect to the L3 ground-truth data.

High-resolution in vivo 4D-OCT fish-eye imaging using 3D-UNet with multi-level residue decoder.

Optical coherence tomography (OCT) allows high-resolution volumetric imaging of biological tissues in vivo. However, 3D-image acquisition often suffers from motion artifacts due to slow frame rates and involuntary and physiological movements of living tissue. To solve these issues, we implement a real-time 4D-OCT system capable of reconstructing near-distortion-free volumetric images based on a deep learning-based reconstruction algorithm. The system initially collects undersampled volumetric images at a high speed and then upsamples the images in real-time by a convolutional neural network (CNN) that generates high-frequency features using a deep learning algorithm. We compare and analyze both dual-2D- and 3D-UNet-based networks for the OCT 3D high-resolution image reconstruction. We refine the network architecture by incorporating multi-level information to accelerate convergence and improve accuracy. The network is optimized by utilizing the 16-bit floating-point precision for network parameters to conserve GPU memory and enhance efficiency. The result shows that the refined and optimized 3D-network is capable of retrieving the tissue structure more precisely and enable real-time 4D-OCT imaging at a rate greater than 10 Hz with a root mean square error (RMSE) of ∼0.03.

Enhancing the Resolution of Satellite Ocean Data Using Discretized Satellite Gridding Neural Networks

Ocean satellite data are often impeded by intrinsic limitations in resolution and accuracy. However, conventional data reconstruction approaches encounter substantial challenges when facing the nonlinear oceanic system and high-resolution fusion of variables. This research presents a Discrete Satellite Gridding Neural Network (DSGNN), a new machine learning method that processes satellite data within a discrete grid framework. By transforming the positional information of grid elements into a standardized vector format, the DSGNN significantly elevates the accuracy and resolution of data fusion through a neural network model. This method’s innovative aspect lies in its discretization and fusion technique, which not only enhances the spatial resolution of oceanic data but also, through the integration of multi-element datasets, better reflects the true physical state of the ocean. A comprehensive analysis of the reconstructed datasets indicates the DSGNN’s consistency and reliability across different seasons and oceanic regions, especially in its adept handling of complex nonlinear interactions and small-scale oceanic features. The DSGNN method has demonstrated exceptional competence in reconstructing global ocean datasets, maintaining small error variance, and achieving high congruence with in situ observations, which is almost equivalent to 1/12° hybrid coordinate ocean model (HYCOM) data. This study offers a novel and potent strategy for the high-resolution reconstruction and fusion of ocean satellite datasets.

Effects of healthy aging on tongue-jaw kinematics during feeding behavior in rhesus macaques.

Several age-related oral health problems have been associated with neurodegenerative diseases such as Alzheimer's Disease (AD), yet how oromotor dysfunction in healthy aging differ from those found in pathological aging is still unknown. This is partly because changes in the cortical and biomechanical ("neuromechanical") control of oromotor behavior in healthy aging are poorly understood. To this end, we investigated the natural feeding behavior of young and aged rhesus macaques (Macaca mulatta) to understand the age-related differences in tongue and jaw kinematics. We tracked tongue and jaw movements in 3D using high-resolution biplanar videoradiography and X-ray Reconstruction of Moving Morphology (XROMM). Older subjects exhibited a reduced stereotypy in tongue movements during chews and a greater lag in tongue movements relative to jaw movements compared to younger subjects. Overall, our findings reveal age-related changes in tongue and jaw kinematics, which may indicate impaired tongue-jaw coordination. Our results have important implications for the discovery of potential neuromechanical biomarkers for early diagnosis of AD.

A time window averaging method to mitigate the impact of shell growth trends on Tridacna δ18O records

Tridacna shells have become valuable archives for high-resolution paleoclimate reconstructions due to their daily-to-annual bands. The δ18O, influenced by temperature and salinity, serves as a key geochemical proxy for Tridacna. Traditionally, linear interpolation has been the go-to method for creating evenly spaced Tridacna δ18O timeseries. However, the accuracy of this method can be influenced by the prominent negative shell growth trends. Given that Tridacna shells exhibit significantly faster growth rates during the juvenile stage than in adulthood, constant sampling intervals may lead to overly dense sampling during the juvenile phase, which could potentially introduce errors in the linearly interpolated Tridacna δ18O record. In this study, we proposed a time window averaging method to mitigate the influence of growth trends in Tridacna δ18O records from the northern South China Sea (SCS). The time window averaging method calculates the average of measured Tridacna δ18O data within a time window, whose length is determined by the sampling resolution for each year and the expected temporal resolution. Pseudo-Tridacna experiments, using the newly developed proxy system model for Tridacna δ18O, were conducted to assess the accuracy of both interpolation and averaging methods. Results indicate that the averaging method outperforms the interpolation method, especially in seasonality estimation. The pseudo-Tridacna experiments recommend employing the averaging method for monthly Tridacna δ18O records when sampling density exceeds ∼20 samples/year. Two real Tridacna δ18O records from the northern SCS demonstrate consistency with pseudo-Tridacna experiments, supporting the efficacy of the averaging method in reducing bias in Tridacna δ18O records associated with growth trends. Our results show the potential of the time window averaging method to improve Tridacna-based paleoclimate reconstructions.

Fast radiance field reconstruction from sparse inputs

Neural Radiance Field (NeRF) has emerged as a powerful method in data-driven 3D reconstruction because of its simplicity and state-of-the-art performance. However, NeRF requires densely captured calibrated images and lengthy training and rendering time to realize high-resolution reconstruction. Thus, we propose a fast radiance field reconstruction method from a sparse set of images with silhouettes. Our approach integrates NeRF with Shape from Silhouette, a traditional 3D reconstruction method that uses silhouette information to fit the shape of an object. To combine NeRF’s implicit representation with Shape from Silhouette’s explicit representation, we propose a novel explicit–implicit radiance field representation consisting of voxel grids with confidence and feature embedding for geometry and a multilayer perceptron network to decode view-dependent color emission for appearance. We propose to make the reconstructed geometry compact by taking advantage of silhouette images, which can avoid the majority of artifacts in sparse input scenarios and speed up training and rendering. We also apply voxel dilating and pruning to refine the geometry prediction. In addition, we impose a total variation regularization on our model to encourage a smooth radiance field. Experiments on the DTU and the NeRF-Synthetic datasets show that our algorithm surpasses the existing baselines in terms of efficiency and accuracy.