Reconstruction Pipeline Research Articles

A high fidelity approach to assembling the complex Borrelia genome

BackgroundBacteria of the Borrelia burgdorferi sensu lato (s.l.) complex can cause Lyme borreliosis. Different B. burgdorferi s.l. genospecies vary in their host and vector associations and human pathogenicity but the genetic basis for these adaptations is unresolved and requires completed and reliable genomes for comparative analyses. The de novo assembly of a complete Borrelia genome is challenging due to the high levels of complexity, represented by a high number of circular and linear plasmids that are dynamic, showing mosaic structure and sequence homology. Previous work demonstrated that even advanced approaches, such as a combination of short-read and long-read data, might lead to incomplete plasmid reconstruction. Here, using recently developed high-fidelity (HiFi) PacBio sequencing, we explored strategies to obtain gap-free, complete and high quality Borrelia genome assemblies. Optimizing genome assembly, quality control and refinement steps, we critically appraised existing techniques to create a workflow that lead to improved genome reconstruction.ResultsDespite the latest available technologies, stand-alone sequencing and assembly methods are insufficient for the generation of complete and high quality Borrelia genome assemblies. We developed a workflow pipeline for the de novo genome assembly for Borrelia using several types of sequence data and incorporating multiple assemblers to recover the complete genome including both circular and linear plasmid sequences.ConclusionOur study demonstrates that, with HiFi data and an ensemble reconstruction pipeline with refinement steps, chromosomal and plasmid sequences can be fully resolved, even for complex genomes such as Borrelia. The presented pipeline may be of interest for the assembly of further complex microbial genomes.

Robust ab initio solution of the cryo-EM reconstruction problem at low resolution with small data sets

Single particle cryo-electron microscopy has become a critical tool in structural biology over the last decade, able to achieve atomic scale resolution in three dimensional models from hundreds of thousands of (noisy) two-dimensional projection views of particles frozen at unknown orientations. This is accomplished by using a suite of software tools to (i) identify particles in large micrographs, (ii) obtain low-resolution reconstructions, (iii) refine those low-resolution structures, and (iv) finally match the obtained electron scattering density to the constituent atoms that make up the macromolecule or macromolecular complex of interest.Here, we focus on the second stage of the reconstruction pipeline: obtaining a low resolution model from picked particle images. Our goal is to create an algorithm that is capable of ab initio reconstruction from small data sets (on the order of a few thousand selected particles). More precisely, we propose an algorithm that is robust, automatic, and fast enough that it can potentially be used to assist in the assessment of particle quality as the data is being generated during the microscopy experiment.

Fast optical coherence tomography angiography image acquisition and reconstruction pipeline for skin application.

Traditional high-quality OCTA images require multi-repeated scans (e.g., 4-8 repeats) in the same position, which may cause the patient to be uncomfortable. We propose a deep-learning-based pipeline that can extract high-quality OCTA images from only two-repeat OCT scans. The performance of the proposed image reconstruction U-Net (IRU-Net) outperforms the state-of-the-art UNet vision transformer and UNet in OCTA image reconstruction from a two-repeat OCT signal. The results demonstrated a mean peak-signal-to-noise ratio increased from 15.7 to 24.2; the mean structural similarity index measure improved from 0.28 to 0.59, while the OCT data acquisition time was reduced from 21 seconds to 3.5 seconds (reduced by 83%).

Forecast of cross-correlation of Chinese Survey Space Telescope cosmic shear tomography with Ali CMB Polarization Telescope cosmic microwave background lensing

ABSTRACT We present a forecast study on the cross-correlation between cosmic shear tomography from the Chinese Survey Space Telescope (CSST) and cosmic microwave background (CMB) lensing from Ali CMB Polarization Telescope (AliCPT-1) in Tibet. The correlated galaxy and CMB lensing signals were generated from Gaussian realizations based on inputted auto and cross-spectra. To account for the error budget, we considered the CMB lensing reconstruction noise based on the AliCPT-1 lensing reconstruction pipeline; shape noise of the galaxy lensing measurement; CSST photo-z error; photo-z bias; intrinsic alignment (IA) effect; and multiplicative bias. The AliCPT-1 CMB lensing mock data were generated according to two experimental stages, namely the ‘4 modules*yr’ and ‘48 modules*yr’ cases. We estimate the cross-spectra in four tomographic bins according to the CSST photo-z distribution in the range of z ∈ [0, 4). After reconstructing the pseudo-cross-spectra from the realizations, we calculate the signal-to-noise ratio (SNR). By combining the four photo-z bins, the total cross-correlation SNR ≈ 15 (AliCPT-1 ‘4 modules*yr’) and SNR ≈ 22 (AliCPT-1 ‘48 modules*yr’). Finally, we study the cosmological application of this cross-correlation signal. Excluding IA in the template fitting would lead to roughly a 0.6σ increment in σ8 due to the negative IA contribution to the galaxy lensing data. For AliCPT-1 first and second stages, the cross-correlation of CSST cosmic shear with CMB lensing gives errors on the clustering amplitude $\sigma _{\sigma _8}=^{+0.043}_{-0.038}$ or $\sigma _{S_8}=\pm 0.031$ and $\sigma _{\sigma _8}=^{+0.030}_{-0.027}$ or $\sigma _{S_8}=\pm 0.018$, respectively.

Cross-Viewpoint Semantic Mapping: Integrating Human and Robot Perspectives for Improved 3D Semantic Reconstruction.

Allocentric semantic 3D maps are highly useful for a variety of human-machine interaction related tasks since egocentric viewpoints can be derived by the machine for the human partner. Class labels and map interpretations, however, may differ or could be missing for the participants due to the different perspectives. Particularly, when considering the viewpoint of a small robot, which significantly differs from the viewpoint of a human. In order to overcome this issue, and to establish common ground, we extend an existing real-time 3D semantic reconstruction pipeline with semantic matching across human and robot viewpoints. We use deep recognition networks, which usually perform well from higher (i.e., human) viewpoints but are inferior from lower viewpoints, such as that of a small robot. We propose several approaches for acquiring semantic labels for images taken from unusual perspectives. We start with a partial 3D semantic reconstruction from the human perspective that we transfer and adapt to the small robot's perspective using superpixel segmentation and the geometry of the surroundings. The quality of the reconstruction is evaluated in the Habitat simulator and a real environment using a robot car with an RGBD camera. We show that the proposed approach provides high-quality semantic segmentation from the robot's perspective, with accuracy comparable to the original one. In addition, we exploit the gained information and improve the recognition performance of the deep network for the lower viewpoints and show that the small robot alone is capable of generating high-quality semantic maps for the human partner. The computations are close to real-time, so the approach enables interactive applications.

From 2D to 3D video conferencing: modular RGB-D capture and reconstruction for interactive natural user representations in immersive extended reality (XR) communication

With recent advancements in Virtual Reality (VR) and Augmented Reality (AR) hardware, many new immersive Extended Reality (XR) applications and services arose. One challenge that remains is to solve the social isolation often felt in these extended reality experiences and to enable a natural multi-user communication with high Social Presence. While a multitude of solutions exist to address this issue with computer-generated “artificial” avatars (based on pre-rendered 3D models), this form of user representation might not be sufficient for conveying a sense of co-presence for many use cases. In particular, for personal communication (for example, with family, doctor, or sales representatives) or for applications requiring photorealistic rendering. One alternative solution is to capture users (and objects) with the help of RGBD sensors to allow real-time photorealistic representations of users. In this paper, we present a complete and modular RGBD capture application and outline the different steps needed to utilize RGBD as means of photorealistic 3D user representations. We outline different capture modalities, as well as individual functional processing blocks, with its advantages and disadvantages. We evaluate our approach in two ways, a technical evaluation of the operation of the different modules and two small-scale user evaluations within integrated applications. The integrated applications present the use of the modular RGBD capture in both augmented reality and virtual reality communication application use cases, tested in realistic real-world settings. Our examples show that the proposed modular capture and reconstruction pipeline allows for easy evaluation and extension of each step of the processing pipeline. Furthermore, it allows parallel code execution, keeping performance overhead and delay low. Finally, our proposed methods show that an integration of 3D photorealistic user representations into existing video communication transmission systems is feasible and allows for new immersive extended reality applications.

Comparison of open‐source three‐dimensional reconstruction pipelines for maize‐root phenotyping

AbstractUnderstanding three‐dimensional (3D) root traits is essential to improve water uptake, increase nitrogen capture, and raise carbon sequestration from the atmosphere. However, quantifying 3D root traits by reconstructing 3D root models for deeper field‐grown roots remains a challenge due to the unknown tradeoff between 3D root‐model quality and 3D root‐trait accuracy. Therefore, we performed two computational experiments. We first compared the 3D model quality generated by five state‐of‐the‐art open‐source 3D model reconstruction pipelines on 12 contrasting genotypes of field‐grown maize roots. These pipelines included COLMAP, COLMAP+PMVS (Patch‐based Multi‐View Stereo), VisualSFM, Meshroom, and OpenMVG+MVE (Multi‐View Environment). The COLMAP pipeline achieved the best performance regarding 3D model quality versus computational time and image number needed. In the second test, we compared the accuracy of 3D root‐trait measurement generated by the Digital Imaging of Root Traits 3D pipeline (DIRT/3D) using COLMAP‐based 3D reconstruction with our current DIRT/3D pipeline that uses a VisualSFM‐based 3D reconstruction on the same dataset of 12 genotypes, with 5–10 replicates per genotype. The results revealed that (1) the average number of images needed to build a denser 3D model was reduced from 3000 to 3600 (DIRT/3D [VisualSFM‐based 3D reconstruction]) to around 360 for computational test 1, and around 600 for computational test 2 (DIRT/3D [COLMAP‐based 3D reconstruction]); (2) denser 3D models helped improve the accuracy of the 3D root‐trait measurement; (3) reducing the number of images can help resolve data storage problems. The updated DIRT/3D (COLMAP‐based 3D reconstruction) pipeline enables quicker image collection without compromising the accuracy of 3D root‐trait measurements.

CATER: Combined Animal Tracking & Environment Reconstruction

Quantifying the behavior of small animals traversing long distances in complex environments is one of the most difficult tracking scenarios for computer vision. Tiny and low-contrast foreground objects have to be localized in cluttered and dynamic scenes as well as trajectories compensated for camera motion and drift in multiple lengthy recordings. We introduce CATER, a novel methodology combining an unsupervised probabilistic detection mechanism with a globally optimized environment reconstruction pipeline enabling precision behavioral quantification in natural environments. Implemented as an easy to use and highly parallelized tool, we show its application to recover fine-scale motion trajectories, registered to a high-resolution image mosaic reconstruction, of naturally foraging desert ants from unconstrained field recordings. By bridging the gap between laboratory and field experiments, we gain previously unknown insights into ant navigation with respect to motivational states, previous experience, and current environments and provide an appearance-agnostic method applicable to study the behavior of a wide range of terrestrial species under realistic conditions.

End-to-end weakly-supervised single-stage multiple 3D hand mesh reconstruction from a single RGB image

In this paper, we consider the challenging task of simultaneously locating and recovering multiple hands from a single 2D image. Previous studies either focus on single hand reconstruction or solve this problem in a multi-stage way. Moreover, the conventional two-stage pipeline firstly detects hand areas, and then estimates 3D hand pose from each cropped patch. To reduce the computational redundancy in preprocessing and feature extraction, for the first time, we propose a concise but efficient single-stage pipeline for multi-hand reconstruction. Specifically, we design a multi-head auto-encoder structure, where each head network shares the same feature map and outputs the hand center, pose and texture, respectively. Besides, we adopt a weakly-supervised scheme to alleviate the burden of expensive 3D real-world data annotations. To this end, we propose a series of losses optimized by a stage-wise training scheme, where a multi-hand dataset with 2D annotations is generated based on the publicly available single hand datasets. In order to further improve the accuracy of the weakly supervised model, we adopt several feature consistency constraints in both single and multiple hand settings. Specifically, the keypoints of each hand estimated from local features should be consistent with the re-projected points predicted from global features. Extensive experiments on public benchmarks including FreiHAND, HO3D, InterHand2.6M and RHD demonstrate that our method outperforms the state-of-the-art model-based methods in both weakly-supervised and fully-supervised manners. The code and models are available at https://github.com/zijinxuxu/SMHR.

Live 4D-OCT denoising with self-supervised deep learning

By providing three-dimensional visualization of tissues and instruments at high resolution, live volumetric optical coherence tomography (4D-OCT) has the potential to revolutionize ophthalmic surgery. However, the necessary imaging speed is accompanied by increased noise levels. A high data rate and the requirement for minimal latency impose major limitations for real-time noise reduction. In this work, we propose a low complexity neural network for denoising, directly incorporated into the image reconstruction pipeline of a microscope-integrated 4D-OCT prototype with an A-scan rate of 1.2 MHz. For this purpose, we trained a blind-spot network on unpaired OCT images using a self-supervised learning approach. With an optimized U-Net, only a few milliseconds of additional latency were introduced. Simultaneously, these architectural adaptations improved the numerical denoising performance compared to the basic setup, outperforming non-local filtering algorithms. Layers and edges of anatomical structures in B-scans were better preserved than with Gaussian filtering despite comparable processing time. By comparing scenes with and without denoising employed, we show that neural networks can be used to improve visual appearance of volumetric renderings in real time. Enhancing the rendering quality is an important step for the clinical acceptance and translation of 4D-OCT as an intra-surgical guidance tool.

Dense light field reconstruction based on epipolar focus spectrum

Existing light field (LF) representations, such as epipolar plane image (EPI) and sub-aperture images, do not consider the structural characteristics across the views, so they usually require additional disparity and spatial structure cues for follow-up tasks. Besides, they have difficulties dealing with occlusions or large disparity scenes. To this end, this paper proposes a novel Epipolar Focus Spectrum (EFS) representation by rearranging the EPI spectrum. Different from the classical EPI representation where an EPI line corresponds to a specific depth, there is a one-to-one mapping from the EFS line to the view. By exploring the EFS sampling task, the analytical function is derived for constructing a non-aliasing EFS. To demonstrate its effectiveness, we develop a trainable EFS-based pipeline for light field reconstruction, where a dense light field can be reconstructed by compensating the missing EFS lines given a sparse light field, yielding promising results with cross-view consistency, especially in the presence of severe occlusion and large disparity. Experimental results on both synthetic and real-world datasets demonstrate the validity and superiority of the proposed method over SOTA methods.

Elucidating yeast glycolytic dynamics at steady state growth and glucose pulses through kinetic metabolic modeling

Microbial cell factories face changing environments during industrial fermentations. Kinetic metabolic models enable the simulation of the dynamic metabolic response to these perturbations, but their development is challenging due to model complexity and experimental data requirements. An example of this is the well-established microbial cell factory Saccharomyces cerevisiae, for which no consensus kinetic model of central metabolism has been developed and implemented in industry. Here, we aim to bring the academic and industrial communities closer to this consensus model. We developed a physiology informed kinetic model of yeast glycolysis connected to central carbon metabolism by including the effect of anabolic reactions precursors, mitochondria and the trehalose cycle. To parametrize such a large model, a parameter estimation pipeline was developed, consisting of a divide and conquer approach, supplemented with regularization and global optimization. Additionally, we show how this first mechanistic description of a growing yeast cell captures experimental dynamics at different growth rates and under a strong glucose perturbation, is robust to parametric uncertainty and explains the contribution of the different pathways in the network. Such a comprehensive model could not have been developed without using steady state and glucose perturbation data sets. The resulting metabolic reconstruction and parameter estimation pipeline can be applied in the future to study other industrially-relevant scenarios. We show this by generating a hybrid CFD-metabolic model to explore intracellular glycolytic dynamics for the first time. The model suggests that all intracellular metabolites oscillate within a physiological range, except carbon storage metabolism, which is sensitive to the extracellular environment.

IBEAT V2.0: a multisite-applicable, deep learning-based pipeline for infant cerebral cortical surface reconstruction.

The human cerebral cortex undergoes dramatic and critical development during early postnatal stages. Benefiting from advances in neuroimaging, many infant brain magnetic resonance imaging (MRI) datasets have been collected from multiple imaging sites with different scanners and imaging protocols for the investigation of normal and abnormal early brain development. However, it is extremely challenging to precisely process and quantify infant brain development with these multisite imaging data because infant brain MRI scans exhibit (a) extremely low and dynamic tissue contrast caused by ongoing myelination and maturation and (b) inter-site data heterogeneity resulting from the use of diverse imaging protocols/scanners. Consequently, existing computational tools and pipelines typically perform poorly on infant MRI data. To address these challenges, we propose a robust, multisite-applicable, infant-tailored computational pipeline that leverages powerful deep learning techniques. The main functionality of the proposed pipeline includes preprocessing, brain skull stripping, tissue segmentation, topology correction, cortical surface reconstruction and measurement. Our pipeline can handle both T1w and T2w structural infant brain MR images well in a wide age range (from birth to 6 years of age) and is effective for different imaging protocols/scanners, despite being trained only on the data from the Baby Connectome Project. Extensive comparisons with existing methods on multisite, multimodal and multi-age datasets demonstrate superior effectiveness, accuracy and robustness of our pipeline. We have maintained a website, iBEAT Cloud, for users to process their images with our pipeline ( http://www.ibeat.cloud ), which has successfully processed over 16,000 infant MRI scans from more than 100 institutions with various imaging protocols/scanners.

Velocity reconstruction with the cosmic microwave background and galaxy surveys

The kinetic Sunyaev Zel'dovich (kSZ) and moving lens effects, secondary contributions to the cosmic microwave background (CMB), carry significant cosmological information due to their dependence on the large-scale peculiar velocity field. Previous work identified a promising means of extracting this cosmological information using a set of quadratic estimators for the radial and transverse components of the velocity field. These estimators are based on the statistically anisotropic components of the cross-correlation between the CMB and a tracer of large scale structure, such as a galaxy redshift survey. In this work, we assess the challenges to the program of velocity reconstruction posed by various foregrounds and systematics in the CMB and galaxy surveys, as well as biases in the quadratic estimators. To do so, we further develop the quadratic estimator formalism and implement a numerical code for computing properly correlated spectra for all the components of the CMB (primary/secondary blackbody components and foregrounds) and a photometric redshift survey, with associated redshift errors, to allow for accurate forecasting. We create a simulation framework for generating realizations of properly correlated CMB maps and redshift binned galaxy number counts, assuming the underlying fields are Gaussian, and use this to validate a velocity reconstruction pipeline and assess map-based systematics such as masking. We highlight the most significant challenges for velocity reconstruction, which include biases associated with: modelling errors, characterization of redshift errors, and coarse graining of cosmological fields on our past light cone. Despite these challenges, the outlook for velocity reconstruction is quite optimistic, and we use our reconstruction pipeline to confirm that these techniques will be feasible with near-term CMB experiments and photometric galaxy redshift surveys.

Real-world CMB lensing quadratic estimator power spectrum response

I describe a method to estimate response matrices of Cosmic Microwave Background (CMB) lensing power spectra estimators to the true sky power under realistic conditions. Applicable to all lensing reconstruction pipelines based on quadratic estimators (QE), it uses a small number of Gaussian CMB Monte-Carlos and specially designed QE's in order to obtain sufficiently accurate matrices with little computational effort. This method may be used to improve the modelling of CMB lensing band-powers by incorporating at least some of the non-idealities encountered in CMB lensing reconstruction. These non-idealities always include masking, and often inhomogeneous filtering, either in the harmonic domain or pixel space. I obtain these matrices for Planck latest lensing reconstructions, and then show that the residual couplings induced by masking explain very well the residual multiplicative bias seen on the Planck simulations, removing the need for an empirical correction.

Toward a realistic in silico abdominal phantom for QSM.

QSM outside the brain has recently gained interest, particularly in the abdominal region. However, the absence of reliable ground truths makes difficult to assess reconstruction algorithms, whose quality is already compromised by additional signal contributions from fat, gases, and different kinds of motion. This work presents a realistic in silico phantom for the development, evaluation and comparison of abdominal QSM reconstruction algorithms. Synthetic susceptibility and maps were generated by segmenting and postprocessing the abdominal 3T MRI data from a healthy volunteer. Susceptibility and values in different tissues/organs were assigned according to literature and experimental values and were also provided with realistic textures. The signal was simulated using as input the synthetic QSM and maps and fat contributions. Three susceptibility scenarios and two acquisition protocols were simulated to compare different reconstruction algorithms. QSM reconstructions show that the phantom allows to identify the main strengths and limitations of the acquisition approaches and reconstruction algorithms, such as in-phase acquisitions, water-fat separation methods, and QSM dipole inversion algorithms. The phantom showed its potential as a ground truth to evaluate and compare reconstruction pipelines and algorithms. The publicly available source code, designed in a modular framework, allows users to easily modify the susceptibility, and TEs, and thus creates different abdominal scenarios.

RECONSTRUÇÃO 3D DE CENAS E OBJETOS A PARTIR DE IMAGENS DIGITAIS

New technologies, such as 3D printers and autonomous cars for instance, originating from advances in Computer Vision and other fields, have been causing an increasingly high interest in robust pipelines for 3D reconstruction of real environments and objects. Through use of 3D reconstruction methods, it is possible to create an application that takes digital photographs of an observed scene as inputs and is capable of obtaining a 3D model that represents it. This model could then be used in a wide range of applications, such as game asset generation, video manipulation with special effects or object replication with the use of a 3D printer, for instance. In this paper, we address, present and discuss methods that concern the different stages of a traditional 3D reconstruction pipeline from digital images only in the implementation and use of computational tools that were capable of generating 3D models that represent a scene observed these images with high fidelity.

Movement Artifact Suppression in Wearable Low-Density and Dry EEG Recordings Using Active Electrodes and Artifact Subspace Reconstruction.

Wearable low-density dry electroencephalogram (EEG) headsets facilitate multidisciplinary applications of brain-activity decoding and brain-triggered interaction for healthy people in real-world scenarios. However, movement artifacts pose a great challenge to their validity in users with naturalistic behaviors (i.e., without highly controlled settings in a laboratory). High-precision, high-density EEG instruments commonly embed an active electrode infrastructure and/or incorporate an auxiliary artifact subspace reconstruction (ASR) pipeline to handle movement artifact interferences. Existing endeavors motivate this study to explore the efficacy of both hardware and software solutions in low-density and dry EEG recordings against non-tethered settings, which are rarely found in the literature. Therefore, this study employed a LEGO-like electrode-holder assembly grid to coordinate three 3-channel system designs (with passive/active dry vs. passive wet electrodes). It also conducted a simultaneous EEG recording while performing an oddball task during treadmill walking, with speeds of 1 and 2 KPH. The quantitative metrics of pre-stimulus noise, signal-to-noise ratio, and inter-subject correlation from the collected event-related potentials of 18 subjects were assessed. Results indicate that while treating a passive-wet system as benchmark, only the active-electrode design more or less rectified movement artifacts for dry electrodes, whereas the ASR pipeline was substantially compromised by limited electrodes. These findings suggest that a lightweight, minimally obtrusive dry EEG headset should at least equip an active-electrode infrastructure to withstand realistic movement artifacts for potentially sustaining its validity and applicability in real-world scenarios.

Polarization Guided HDR Reconstruction via Pixel-Wise Depolarization.

Taking photos with digital cameras often accompanies saturated pixels due to their limited dynamic range, and it is far too ill-posed to restore them. Capturing multiple low dynamic range images with bracketed exposures can make the problem less ill-posed, however, it is prone to ghosting artifacts caused by spatial misalignment among images. A polarization camera can capture four spatially-aligned and temporally-synchronized polarized images with different polarizer angles in a single shot, which can be used for ghost-free high dynamic range (HDR) reconstruction. However, real-world scenarios are still challenging since existing polarization-based HDR reconstruction methods treat all pixels in the same manner and only utilize the spatially-variant exposures of the polarized images (without fully exploiting the degree of polarization (DoP) and the angle of polarization (AoP) of the incoming light to the sensor, which encode abundant structural and contextual information of the scene) to handle the problem still in an ill-posed manner. In this paper, we propose a pixel-wise depolarization strategy to solve the polarization guided HDR reconstruction problem, by classifying the pixels based on their levels of ill-posedness in HDR reconstruction procedure and applying different solutions to different classes. To utilize the strategy with better generalization ability and higher robustness, we propose a network-physics-hybrid polarization-based HDR reconstruction pipeline along with a neural network tailored to it, fully exploiting the DoP and AoP. Experimental results show that our approach achieves state-of-the-art performance on both synthetic and real-world images.

Toward 3D scene reconstruction from locally scale-aligned monocular video depth

Monocular depth estimation methods have achieved excellent robustness on diverse scenes, usually by predicting affine-invariant depth, up to an unknown scale and shift, rather than metric depth in that it is much easier to collect large-scale affine-invariant depth training data. However, in some video-based scenarios such as video depth estimation and 3D scene reconstruction, the unknown scale and shift residing in per-frame prediction may cause the predicted depth to be inconsistent. To tackle this problem, we propose a locally weighted linear regression method to recover the scale and shift map with very sparse anchor points, which ensures the consistency along consecutive frames. Extensive experiments show that our method can drop the Rel error (relative error) of existing state-of-the-art approaches significantly over several zero-shot benchmarks. Besides, we merge 6.3 million RGBD images to train robust depth models. By locally recovering scale and shift, our produced ResNet50-backbone model even outperforms the state-of-the-art DPT ViT-Large model. Combined with geometry-based reconstruction methods, we formulate a new dense 3D scene reconstruction pipeline, which benefits from both the scale consistency of sparse points and the robustness of monocular methods. By performing simple per-frame prediction over a video, the accurate 3D scene geometry can be recovered.