Extra Storage Research Articles

FastComposer: Tuning-Free Multi-subject Image Generation with Localized Attention

AbstractDiffusion models excel at text-to-image generation, especially in subject-driven generation for personalized images. However, existing methods are inefficient due to the subject-specific fine-tuning, which is computationally intensive and hampers efficient deployment. Moreover, existing methods struggle with multi-subject generation as they often blend identity among subjects. We present FastComposer which enables efficient, personalized, multi-subject text-to-image generation without fine-tuning. FastComposer uses subject embeddings extracted by an image encoder to augment the generic text conditioning in diffusion models, enabling personalized image generation based on subject images and textual instructions with only forward passes. To address the identity blending problem in the multi-subject generation, FastComposer proposes cross-attention localization supervision during training, enforcing the attention of reference subjects localized to the correct regions in the target images. Naively conditioning on subject embeddings results in subject overfitting. FastComposer proposes delayed subject conditioning in the denoising step to maintain both identity and editability in subject-driven image generation. FastComposer generates images of multiple unseen individuals with different styles, actions, and contexts. It achieves 300$$\times $$ × –2500$$\times $$ × speedup compared to fine-tuning-based methods and requires zero extra storage for new subjects. FastComposer paves the way for efficient, personalized, and high-quality multi-subject image creation. Code, model, and dataset are available here (https://github.com/mit-han-lab/fastcomposer).

Nanosheet-interwoven structures and ion-electron decoupling storage enable Fe1-xS fast ion transport in Li+/Na+/K+ batteries

Fe1-xS, known for its high theoretical capacity, abundant resources, and intrinsic safety, has become a focal point as a universal anode for Li+/Na+/K+ batteries. However, its fast-charging capability is unsatisfactory due to sluggish ion transport rate and low electrical conductivity. Furthermore, its Li+/Na+/K+ storage mechanisms are still unclear. Here, we fabricate a single-crystal Fe1-xS/N-doped carbon composite nanosheet interwoven structure, in which N-doped carbon layers onto surface of Fe1-xS nanosheets ameliorate the electrical conduction and the interwoven nanosheets form open pore channels that favor permeation of electrolytes to boost ion transport. In-situ magnetometry reveals that ion-electron decoupling storage and transport occur in two-phase composites of Fe/Li2S, Fe/Na2S, and Fe/K2S, in which Fe phase stores and transports electrons and sulfide phase stores and transports ions in a space-charge form, resulting in extra ion storage and fast ion transport. Consequently, the nanosheet interwoven structure delivers high capacities (1320.1/652.2/350.6 mAh g−1), outstanding fast-charging performances (679.6/295.4/106.4 mAh g−1 at 20 A g−1), and long cycling life over 5000 cycles as Li+/Na+/K+ battery anodes, respectively

Elastic MSM: A Fast, Elastic and Modular Preprocessing Technique for Multi-Scalar Multiplication Algorithm on GPUs

Zero-knowledge proof (ZKP) is a cryptographic primitive that enables a prover to convince a verifier that a statement is true, without revealing any other information beyond the correctness of the statement itself. Due to its powerful capabilities, its most practical type, called zero-knowledge Succinct Non-interactive ARgument of Knowledge (zkSNARK), has been widely deployed in various privacypreserving applications such as cryptocurrencies and verifiable computation. Although state-of-the-art zkSNARKs are highly efficient for the verifier, the computational overhead for the prover is still orders of magnitude too high to warrant use in many applications. This overhead arises from several time-consuming operations, including large-scale matrix-vector multiplication (MUL), number-theoretic transform (NTT), and especially the multi-scalar multiplication (MSM) which constitutes the largest proportion. Therefore, further efficiency improvements are needed.In this paper, we focus on comprehensive optimization of running time and storage space required by the MSM algorithm on GPUs. Specifically, we propose a novel, modular and adaptive parameter configuration technique—elastic MSM to enable us to adjust the scale of MSM according to our own wishes by performing a corresponding amount of preprocessing. This technique enables us to fully unleash the potential of various efficient parallel MSM algorithms. We have implemented and tested elastic MSM over three prevailing parallel Pippenger algorithms on GPUs. Across various preprocessing space limitations (across various MSM scales), our constructions achieve up to about 1.90×, 1.08× and 1.36× (2.58×, 1.39× and 1.91×) speedup versus three state-of-the-art parallel Pippenger algorithms on GPUs, respectively.From another perspective, elastic MSM could also be regarded as a preprocessing technique over the well-known Pippenger algorithm, which is modular and could be used to accelerate almost all the most advanced parallel Pippenger algorithms on GPUs. Meanwhile, elastic MSM provides an adaptive trade-off between the running time and the extra storage space needed by parallel Pippenger algorithms on GPUs. This is the first preprocessing technique to retain the improved MSM computation brought by preprocessing under varying storage space limitations. Specifically, across various preprocessing space limitations (across various MSM scales), our constructions achieve up to about 192× and 223× (159× and 174×) speedup versus two state-ofthe- art preprocessing parallel Pippenger algorithms on GPUs, respectively.

Multi-dimensional carbon modification strategy for lithium-ion battery anodes: Carbon-covered TiP2O7 supported by 3D carbon skeleton

Carbon coating is a beneficial method for improving surface capacitive behavior, electron transfer kinetics and cyclic stability of three-dimensional lattice lithium-embedded anode materials. Herein, we adopt one-step pyrolysis to achieve carbon-covered TiP2O7 supported by a three-dimensional carbon skeleton utilizing phytic acid and glucose as carbon sources. The two-dimensional amorphous carbon layer on TiP2O7 surface improves the initial Coulombic efficiency and rate capacity by inhibiting the irreversible embedding of lithium ions and reducing the interfacial charge transfer impedance. The three-dimensional carbon skeleton not only forms a conductive pathway between the carbon-covered TiP2O7, but also its large specific surface area, randomly oriented and defective carbon layer provide extra lithium storage sites and capacitive contribution. In addition, the specific capacity of the TiP2O7/C composite is still as high as 314.5 mA h g−1 after 1000 cycles at 1 A g−1. A TiP2O7/C//LiFePO4 full cell delivers a high energy density of 273.96 Wh kg−1 at 0.153 kW kg−1, and retains 114.43 Wh kg−1 even at the power density of 4.086 kW kg−1. The multi-dimensional carbon modification strategy has improved the electrochemical performance of TiP2O7-based anode materials, providing a model for synthesizing other high-performance anode materials.

Renewable Energy Prediction Using Different ML Algorithms

Abstract: The global community is grappling with numerous obstacles in fulfilling its energy requirements, stemming from the limited availability of fossil fuel reserves. It is, at present, rapidly shifting its focus towards Renewable Energy Sources to overcome the limitations of fossil fuels and generate environmentally friendly energy. Earlier, their unpredictable nature made them less appealing, but recent progress in Artificial Intelligence (AI) and Machine Learning (ML) has allowed for accurate forecasts of their energy production. With a well-planned approach to balancing energy loads, renewable energy sources can satisfy energy needs without the necessity for extra storage. In this piece, we conducted a machine learning analysis on a dataset obtained from the Kaggle Platform by utilizing various algorithms, such as Random Forest (RF), Linear Support Vector Regression (SVM), Decision Tree (DT), and Linear Regression (LR).

Monodisperse Manganese-Vanadium-Oxo Clusters with Extraordinary Lithium Storage.

Although possessing well-defined nanostructures and excellent multi-electron redox properties, polyoxometalate clusters have poor intrinsic electrical conductivity and are prone to aggregation due to large surface energy, which makes them difficult to be fully utilized when applying as electrode materials for lithium-ion batteries. In this paper, monodisperse K7MnV13O38 (MnV13) clusters are achieved by rationally utilizing nano-sized high conductive carbon dots (CDs) as stabilizers. Benefiting from the fully exposed redox sites of MnV13 clusters (high utilization rate) and sufficient interfaces with carbon dots (extra interfacial energy storage), the optimized MnV13/10CDs anode delivers a high discharge capacity up to 1348 mAh g-1 at a current density of 0.1 A g-1 and exhibits superb rate/cycling capabilities. Density functional theory (DFT) calculations verify that ionic archway channels are formed between MnV13 and CDs, eliminating the bandgap and greatly improving the electron/ion conductivity of MnV13 and CDs. This paper paves a brand-new way for synthesis of monodisperse clusters and maximization of extra interfacial energy storage.

Selection of energy storage systems for a special purpose rail vehicle based on simulation analysis

The issue of power supply to electric rail vehicles leads to a separation of the rail network into electrified and unelectrified portions, where the sections lacking electrification exclude the operation of electric rail vehicles powered from the overhead lines. The potential solution to this problem was found in adding energy storage systems to electric rail vehicles, to allow them some range of travel beyond the electrified lines. A simulation analysis of a special purpose rail vehicle travelling across a non-electrified section of railway line was conducted to assess the energy consumption rate and the necessary energy storage capacity. Three energy storage solutions were simulated, showing the travel range they can provide, with the aim of finding the lowest battery capacity solution that would still allow the vehicle to safely complete the simulated drive. The final selection of energy storage system capacity was done based on the assumed expected range outside electrified railway weighed against the mass and cost of the extra energy storage system added to the vehicle. For a vehicle with a mass of 65 tons a battery system with a capacity of 600 Ah was found to be sufficient.

A Comparative Study on Electrochemical Performance of Single versus Dual Networks in Lithium Metal/Polysulfide-Polyoxide Co-Network/Lithium Titanium Oxide Cathode

The present article introduces a strategy for controlling oxidation and reduction reactions within polymer electrolyte membrane (PEM) networks as a means of enhancing storage capacity through the complexation of dissociated lithium cations with multifunctional groups of the polymer network. Specifically, co-polymer networks based on polysulfide (PS) and polyoxide (PO) precursors, photo-cured in the presence of succinonitrile (SCN) and lithium bis(trifluoro methane sulfonyl imide) (LiTFSI) salt, exhibited ionic conductivity on the order of mid 10−4 S/cm at ambient temperature in the 30/35/35 (weight %) composition. Lithium titanate (LTO, Li4Ti5O12) electrode was chosen as an anode (i.e., a potential source of Li ions) against lithium iron phosphate (LFP, LiFePO4) cathode in conjunction with polysulfide-co-polyoxide dual polyelectrolyte networks to control viscosity for 3D printability on conformal surfaces of drone and aeronautic vehicles. It was found that the PS-co-PO dual network-based polymer electrolyte containing SCN plasticizer and LiTFSI salt exhibited extra storage capacity (i.e., specific capacity of 44 mAh/g) with the overall specific capacity of 170 mAh/g (i.e., for the combined LTO electrode and PEM) initially that stabilized at 153 mAh/g after 50th cycles with a reasonable capacity retention of over 90% and Coulombic efficiency of over 99%. Of particular interest is the observation of the improved electrochemical performance of the polysulfide-co-polyoxide electrolyte dual-network relative to that of the polyoxide electrolyte single-network.

Solar-integrated binary chemical cracking of heavy oil for efficient high-order fuel transformation and extra hydrogen storage

Solar energy is an attractive alternative to fossil fuels (electricity and coke) for petroleum products generation. A solar-integrated binary chemical cracking of heavy oil system integrates solar energy into oil processing for carbon neutrality，in which pyrolysis and electrolysis benefit heavy oil conversion and hydrogen storage. This system has been studied in terms of solar chemical engineering, electrolyte engineering and electrode engineering. The results show improved hydrogen yield and cracking rate of heavy oil and a significant reduction in the operating temperature compared to catalytic cracking. The performance of molten salt electrolytes is as follows: binary NaOH–KOH > ternary KCl–LiCl + NaOH > ternary KCl–LiCl + NaHCO3. In electrode engineering, the electro-catalytic electrodes are in an order of Inconel 718> Ni > Incoloy 825 electrodes in the cracking rate, Ni > Inconel 718>Incoloy 825 electrodes in the hydrogen yield, and Incoloy 825>Inconel 718>Ni electrodes in the ability, for sustainable and stable efficient production.

In-situ capture defects through molecule grafting assisted in coal-based hard carbon anode for sodium-ion batteries

Sodium-ion batteries is not essential but also potential to support energy storage system due to the low cost. It's worth noting that low price and high carbon production helps coal extraordinary promise as hard carbon precursor. However, low initial coulombic efficiency (ICE) and practical Na+ storage specific capacity barrier their application owing to lots of defects. Herein, a grafting method is proposed to decrease defects, introducing glucose with simple one step carbonization process. Utilizing the reaction between glucose and coal reduces the defects, which generates a high ICE with less irreversible sodium ion consumption. Meanwhile the expanded carbon layer spaces enable the fast sodium ions transfer, leading to an optimized sodium storage performance. The grafted glucose obviously decreases the defects during the carbonization process and contributes extra sodium storage capacity. The optimal CG0.1 exhibits higher reversible capacity of 293 mAh∙g−1 and ICE of 84 % than pristine coal pyrolytic carbon (250 mAh∙g−1 and 79 %). Besides, it also delivers outstanding capacity retention with 95.3 % and 95.8 % after 1000 cycles at 0.3 A∙g−1 (240 mAh∙g−1) and at 1 A∙g−1 (207 mAh∙g−1).

Performance evaluation of lung sounds classification using deep learning under variable parameters

It is desired to apply deep learning models (DLMs) to assist physicians in distinguishing abnormal/normal lung sounds as quickly as possible. The performance of DLMs depends on feature-related and model-related parameters heavily. In this paper, the relationship between performance and feature-related parameters of a DLM, i.e., convolutional neural network (CNN) is analyzed through experiments. ICBHI 2017 is selected as the lung sounds dataset. The sensitivity analysis of classification performance of the DLM on three parameters, i.e., the length of lung sounds frame, overlap percentage (OP) of successive frames and feature type, is performed. An augmented and balanced dataset is acquired by the way of white noise addition, time stretching and pitch shifting. The spectrogram and mel frequency cepstrum coefficients of lung sounds are used as features to the CNN, respectively. The results of training and test show that there exists significant difference on performance among various parameter combinations. The parameter OP is performance sensitive. The higher OP, the better performance. It is concluded that for fixed sampling frequency 8 kHz, frame size 128, OP 75% and spectrogram feature is optimum under which the performance is relatively better and no extra computation or storage resources are required.

Examine Heuristic Data Lake Management Using AWS: A Big Data Handling Approach

In this era of technology, the most valued asset can be ‘Data’. With the increasing number of data, the value of it keeps increasing. Data storage and data manipulate for to achieve some particular goals or business requirements increasing in number and storing it has become a complex and tedious task. With the use of some advanced technologies like hadoop, it simplified the data storing process, but due to rapid development and excessive use of AI and ML, tons of data is collected. The quintessence is to ascertain an extra cost effective storage alternative. This paper provides with an effective solution to store data over the cloud with numerous benefits over traditional data storage methods by developing a data lake using AWS a Cost Effective Data Lake Management algorithm (CEDLMA). Furthermore, the functionalities of data lake include managing and storing sorted as well as unsorted data, gathering various analytics from the data lake as per business requirements. Proposed work is evaluated with AWS’s IAM and S3 services.

Trend-Aware Supervision: On Learning Invariance for Semi-supervised Facial Action Unit Intensity Estimation

With the increasing need for facial behavior analysis, semi-supervised AU intensity estimation using only keyframe annotations has emerged as a practical and effective solution to relieve the burden of annotation. However, the lack of annotations makes the spurious correlation problem caused by AU co-occurrences and subject variation much more prominent, leading to non-robust intensity estimation that is entangled among AUs and biased among subjects. We observe that trend information inherent in keyframe annotations could act as extra supervision and raising the awareness of AU-specific facial appearance changing trends during training is the key to learning invariant AU-specific features. To this end, we propose Trend-AwareSupervision (TAS), which pursues three kinds of trend awareness, including intra-trend ranking awareness, intra-trend speed awareness, and inter-trend subject awareness. TAS alleviates the spurious correlation problem by raising trend awareness during training to learn AU-specific features that represent the corresponding facial appearance changes, to achieve intensity estimation invariance. Experiments conducted on two commonly used AU benchmark datasets, BP4D and DISFA, show the effectiveness of each kind of awareness. And under trend-aware supervision, the performance can be improved without extra computational or storage costs during inference.

Pressure-Induced Defects and Reduced Size Endow TiO2 with High Capacity over 20 000 Cycles and Excellent Fast-Charging Performance in Sodium Ion Batteries.

Anatase TiO2 as sodium-ion-battery anode has attracted increased attention because of its low volume change and good safety. However, low capacity and poor rate performance caused by low electrical conductivity and slow ion diffusion greatly impede its practical applications. Here, a bi-solvent enhanced pressure strategy that induces defects (oxygen vacancies) into TiO2 via N doping and reduces its size by using mutual-solvent ethanol and dopant dimethylformamide as pressure-increased reagent of tetrabutyl orthotitanate tetramer is proposed to fabricate N-doped TiO2/C nanocomposites. The induced defects can increase ion storage sites, improve electrical conductivity, and decrease bandgap and ion diffuse energy barrier of TiO2. The size reduction increases contact interfaces between TiO2 and C and shortens ion diffuse distance, thus increasing extra ion storage sites and boosting ion diffusion rate of TiO2. The N-doped TiO2 possesses highly stable crystal structure with a slightly increase of 0.86% in crystal lattice spacing and 3.2% in particle size after fully sodiation. Consequently, as a sodium-ion battery anode, the nanocomposite delivers high capacity and superior rate capability along with ultralong cycling life. This work proposes a novel pressure-induced synthesis strategy that provides unique guidance for designing TiO2-based anode materials with high capacity and excellent fast-charging capability.

Yolk‐Shell Structure and Spin‐Polarized Surface Capacitance Enable FeS Stable and Fast Ion Transport in Sodium‐Ion Batteries

AbstractIron sulfide (FeS) has been extensively studied as sodium‐ion battery anodes due to its high theoretical capacity (609 mAh g−1), but its large volume expansion and low electrical conductivity result in unsatisfactory cycling life and poor rate performance. Moreover, the sodium ion storage mechanism of FeS at a voltage range of 0.01–1 V involving conversion reactions and subsequent ion storage process is unclear yet. Here, the study proposes a vapor‐pressure induced synthesis route to fabricate FeS/C yolk‐shell structure that ultrathin carbon layers coat on the surface of FeS nanosheets, which can accommodate volume expansion of FeS during sodiation observed via in situ transmission electron microscope and improve its electrical conductivity. Remarkably, an in situ magnetometry reveals that vast spin‐polarized electrons can be injected into superparamagnetic Fe nanoparticles (≈3 nm) formed during conversion reaction to induce evolution of electrode magnetization between 0.01 and 1 V, during which spin‐polarized surface capacitance effect occurs at Fe/Na2S interfaces to increase extra ion storage and boost ion transport stably. Consequently, the FeS/C yolk‐shell nanosheets deliver a high reversible capacity of 664.9 mAh g−1 at 0.1 A g−1, and 300.4 mAh g−1 after 10 000 cycles at 10 A g−1 with a capacity retention of 81.1%.

Response of maize yield to changes in soil organic matter in a Swedish long‐term experiment

AbstractAgricultural practices that lead to soil carbon sequestration may be a win–win strategy for mitigating global warming and improving soil fertility and resource use efficiency. The mechanisms through which soil organic carbon (SOC) concentration affects crop yields are numerous but difficult to separate. The objective of this study was to disentangle these processes and estimate to what extent the yield response to SOC is mainly driven by changes in physical or biochemical properties and processes. This was achieved by analysing the response of yields in continuous maize to SOC concentrations during 20 years (2000–2019), which had evolved in 14 experimental treatments in a Swedish long‐term field experiment at Ultuna since 1956, ranging from 0.94% to 3.65% in the topsoil (0–20 cm). Average maize yields during this period varied between 1.9 and 8.4 Mg dry mass per hectare in the different treatments. The treatments comprise applications of different mineral nitrogen (N) fertilizers and organic amendments and combinations thereof. Our analysis showed that maize yield in the treatments that were not severely limited by nitrogen supply or soil acidity increased by 16% for each percentage unit increase in SOC. We applied the widely used concept of critical N concentration in plant biomass to diagnose the N status in maize in the different treatments (N nutrition index [NNI]) and parameterized a response function between yield and pH (RpH). Dry soil bulk density (BD) was used as a proxy for soil physical properties. These three variables NNI, RpH and BD explained 95% of the variation in maize yields among treatments. Further analysis of the relationship between BD, SOC and plant available water capacity revealed that about two thirds of the yield increases in response to SOC change could be ascribed to associated changes in soil physical properties. Our analysis suggests that the extra storage capacity of water, which increased by up to 15 mm in the topsoil for each unit percentage increase in SOC, was the main driver for the observed yield responses. We conclude that measures for increasing SOC in soils most likely are an effective adaptation strategy for reducing the risk of crop damage during dry spells, which probably are becoming more frequent in the future due to climate change, even in relatively humid climates as in Sweden.

Encapsulation of Titanium Disulfide into MOF-Derived N,S-Doped Carbon Nanotablets Toward Suppressed Shuttle Effect and Enhanced Sodium Storage Performance.

Titanium disulfide (TiS2) is a promising anode material for sodium-ion batteries due to its high theoretical capacity, but it suffers from severe volume variation and shuttle effect of the intermediate polysulfides. To overcome the drawbacks, herein the successful fabrication of TiS2@N,S-codoped C (denoted as TiS2@NSC) through a chemical vapor reaction between Ti-based metal-organic framework (NH2-MIL-125) and carbon disulfide (CS2) is demonstrated. The C─N bonds enhance the electronic/ionic conductivity of the TiS2@NSC electrode, while the C─S bonds provide extra sodium storage capacity, and both polar bonds synergistically suppress the shuttle effect of polysulfides. Consequently, the TiS2@NSC electrode demonstrates outstanding cycling stability and rate performance, delivering reversible capacities of 418/392mAhg-1 after 1000 cycles at 2/5Ag-1. Ex situ X-ray photoelectron spectroscopy and transmission electron microscope analyses reveal that TiS2 undergoes an intercalation-conversion ion storage mechanism with the generation of metallic Ti in a deeper sodiation state, and the pristine hexagonal TiS2 is electrochemically transformed into cubic rock-salt TiS2 as a reversible phase with enhanced reaction kinetics upon sodiation/desodiation cycling. The strategy to encapsulate TiS2 in N,S-codoped porous carbon matrices efficiently realizes superior conductivity and physical/chemical confinement of the soluble polysulfides, which can be generally applied for the rational design of advanced electrodes.

Enhancing System Reliability and Battery Longevity through Integrated Energy Sources and Algorithmic Optimization

This research examines the enhancement of low power electronic systems, such as low power wide area networks (LPWANs), by incorporating a mix of renewable energy sources like solar and wind in China. It addresses the challenges posed by the variability of these sources due to climatic and temporal factors by merging diverse energy harvesting (EH) methods to bolster power stability and availability. The study assesses the most effective combination of solar, wind, and rain EH for steady power supply. A novel algorithm is introduced for energy input management based on real‐time resource availability, with the goal of prolonging battery life by avoiding complete charge or discharge cycles. The feasibility of this multisource approach in possibly reducing the need for extra energy storage is evaluated. The investigation includes simulations using Chinese weather data to explore different combinations of energy sources. A real‐life system utilizing solar and wind energy, guided by the developed algorithm, was also implemented for empirical comparison. Findings suggest that solar and wind energies have a higher power yield than rain harvesting, with actual wind energy collection often falling short of simulated forecasts. The solar–wind energy mix achieves a 99% system availability and facilitates a more compact design owing to their high power density. Although the input‐switching algorithm lessens the frequency of complete battery drainage, a small degree of energy storage is still essential for maintaining system reliability. This integrated EH method, compared to single‐source options, allows for a smaller energy storage requirement. Moreover, the algorithm limits battery charging to 80%, significantly enhancing battery lifespan.

Lath decomposition and dispersion induced by heterostructure deformation and associated interactions in a lamellae metastable-β Ti alloy during subtransus processing

The mechanisms of lath decomposition and dispersion were hitherto rarely explained due to limited understanding into the heterostructure deformation between two phases and associated interactions. In this work, the lath decomposition and dispersion were systematically investigated in a wide subtransus processing range of 0.001 s−1–1 s−1 and 745 °C–820 °C with the consideration into the heterostructure deformation and associated interactions. The relationship between flowing behavior and microstructure evolution were discussed. Results indicated that the initial buckling of α laths caused high peak strength. The rotation and twist of α laths are helpful to decrease the flow stress, but it caused serious extra deformation and dislocation storage in the β matrix. The lath decomposition dominated by the ablation was greatly contributed by the adiabatic temperature rise (ATR) near boundaries with high energies. The α laths could rotate under low strain rates, and numerous residual laths (RLs) could be obtained. The increment in strain rates to 0.1 s−1 resulted in earlier lath decomposition and softening, as well as to disperse αp particles and homogenize the β matrix. But an overhigh strain rate of ∼1 s−1 caused adiabatic shear bands. The low temperatures are unfavorable to the lath decomposition, while over-high temperatures diminished the extra deformation and ATR. On this basis, a deformation mechanism map for the subtransus processing of lamellae Ti-55531 alloy was established. This investigation sheds deeper insight into the tailoring of microstructure and prediction of flowing behavior for the subtransus processing of lamellae metastable-β Ti alloys.

MIPS-Fusion: Multi-Implicit-Submaps for Scalable and Robust Online Neural RGB-D Reconstruction

We introduce MIPS-Fusion, a robust and scalable online RGB-D reconstruction method based on a novel neural implicit representation - multi-implicit-submap. Different from existing neural RGB-D reconstruction methods lacking either flexibility with a single neural map or scalability due to extra storage of feature grids, we propose a pure neural representation tackling both difficulties with a divide-and-conquer design. In our method, neural submaps are incrementally allocated alongside the scanning trajectory and efficiently learned with local neural bundle adjustments. The submaps can be refined individually in a back-end optimization and optimized jointly to realize submap-level loop closure. Meanwhile, we propose a hybrid tracking approach combining randomized and gradient-based pose optimizations. For the first time, randomized optimization is made possible in neural tracking with several key designs to the learning process, enabling efficient and robust tracking even under fast camera motions. The extensive evaluation demonstrates that our method attains higher reconstruction quality than the state of the arts for large-scale scenes and under fast camera motions.