A Torch Without Light: Low-Light Imaging for Mobile Phones

Photography used to be a hobby that required equipment such as a professional camera. Today, photography has evolved to be a daily activity conducted on an unprecedented scale due to the adoption of camera into smartphones. Mobile phone cameras are on the way to completely replace other forms of camera due to their portability and quality. Millions of images are captured on mobile devices across the globe. These images are clear and crisp. But all these images are captured in daylight. Images taken in low illumination essentially turn out to be too dark to be comprehensible. Research shows that current solutions to this problem work for dim to moderate light level but fail in extreme low light. There are certain problems involved with these techniques. Firstly, image denoising relies on image priors limiting the situations on what it will work on. Other deep learning techniques work on synthetic data and cannot be proficient on real data. Secondly, Low light image enhancement assumes that images already contain a good representation of scene content. This paper proposes to capture low illumination images and transform them to high quality images using end to end fully convolutional neural network trained on our data set of raw images shot in low aperture and their corresponding high aperture raw images. As an outcome, we will be able to transform images to high quality and identify objects.

Download Full-text

TOWARDS LEARNING LOW-LIGHT INDOOR SEMANTIC SEGMENTATION WITH ILLUMINATION-INVARIANT FEATURES

ISPRS - International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences ◽

10.5194/isprs-archives-xliii-b2-2021-427-2021 ◽

2021 ◽

Vol XLIII-B2-2021 ◽

pp. 427-432

Author(s):

N. Zhang ◽

F. Nex ◽

N. Kerle ◽

G. Vosselman

Keyword(s):

Synthetic Data ◽

Real Data ◽

Semantic Segmentation ◽

Data Sets ◽

Low Light ◽

Data Set ◽

Illumination Invariant ◽

Invariant Features ◽

Illumination Changes ◽

Segmentation Models

Abstract. Semantic segmentation models are often affected by illumination changes, and fail to predict correct labels. Although there has been a lot of research on indoor semantic segmentation, it has not been studied in low-light environments. In this paper we propose a new framework, LISU, for Low-light Indoor Scene Understanding. We first decompose the low-light images into reflectance and illumination components, and then jointly learn reflectance restoration and semantic segmentation. To train and evaluate the proposed framework, we propose a new data set, namely LLRGBD, which consists of a large synthetic low-light indoor data set (LLRGBD-synthetic) and a small real data set (LLRGBD-real). The experimental results show that the illumination-invariant features effectively improve the performance of semantic segmentation. Compared with the baseline model, the mIoU of the proposed LISU framework has increased by 11.5%. In addition, pre-training on our synthetic data set increases the mIoU by 7.2%. Our data sets and models are available on our project website.

Download Full-text

Regularization and datuming of seismic data by weighted, damped least squares

Geophysics ◽

10.1190/1.2235616 ◽

2006 ◽

Vol 71 (5) ◽

pp. U67-U76 ◽

Cited By ~ 7

Author(s):

Robert J. Ferguson

Keyword(s):

Least Squares ◽

Seismic Data ◽

Synthetic Data ◽

Real Data ◽

Structural Data ◽

Irregular Sampling ◽

Data Set ◽

Near Surface ◽

Damped Least Squares ◽

Wavefield Extrapolation

The possibility of improving regularization/datuming of seismic data is investigated by treating wavefield extrapolation as an inversion problem. Weighted, damped least squares is then used to produce the regularized/datumed wavefield. Regularization/datuming is extremely costly because of computing the Hessian, so an efficient approximation is introduced. Approximation is achieved by computing a limited number of diagonals in the operators involved. Real and synthetic data examples demonstrate the utility of this approach. For synthetic data, regularization/datuming is demonstrated for large extrapolation distances using a highly irregular recording array. Without approximation, regularization/datuming returns a regularized wavefield with reduced operator artifacts when compared to a nonregularizing method such as generalized phase shift plus interpolation (PSPI). Approximate regularization/datuming returns a regularized wavefield for approximately two orders of magnitude less in cost; but it is dip limited, though in a controllable way, compared to the full method. The Foothills structural data set, a freely available data set from the Rocky Mountains of Canada, demonstrates application to real data. The data have highly irregular sampling along the shot coordinate, and they suffer from significant near-surface effects. Approximate regularization/datuming returns common receiver data that are superior in appearance compared to conventional datuming.

Download Full-text

Bayesian inversion of magnetotelluric data considering dimensionality discrepancies

Geophysical Journal International ◽

10.1093/gji/ggaa391 ◽

2020 ◽

Vol 223 (3) ◽

pp. 1565-1583

Author(s):

Hoël Seillé ◽

Gerhard Visser

Keyword(s):

Learning Algorithm ◽

Synthetic Data ◽

Real Data ◽

Error Model ◽

Bayesian Inversion ◽

Magnetotelluric Data ◽

Data Set ◽

Likelihood Functions ◽

Training Images ◽

Phase Tensor

SUMMARY Bayesian inversion of magnetotelluric (MT) data is a powerful but computationally expensive approach to estimate the subsurface electrical conductivity distribution and associated uncertainty. Approximating the Earth subsurface with 1-D physics considerably speeds-up calculation of the forward problem, making the Bayesian approach tractable, but can lead to biased results when the assumption is violated. We propose a methodology to quantitatively compensate for the bias caused by the 1-D Earth assumption within a 1-D trans-dimensional Markov chain Monte Carlo sampler. Our approach determines site-specific likelihood functions which are calculated using a dimensionality discrepancy error model derived by a machine learning algorithm trained on a set of synthetic 3-D conductivity training images. This is achieved by exploiting known geometrical dimensional properties of the MT phase tensor. A complex synthetic model which mimics a sedimentary basin environment is used to illustrate the ability of our workflow to reliably estimate uncertainty in the inversion results, even in presence of strong 2-D and 3-D effects. Using this dimensionality discrepancy error model we demonstrate that on this synthetic data set the use of our workflow performs better in 80 per cent of the cases compared to the existing practice of using constant errors. Finally, our workflow is benchmarked against real data acquired in Queensland, Australia, and shows its ability to detect the depth to basement accurately.

Download Full-text

Supervised Image Classification by Scattering Transform with Application to Weed Detection in Culture Crops of High Density

Remote Sensing ◽

10.3390/rs11030249 ◽

2019 ◽

Vol 11 (3) ◽

pp. 249 ◽

Cited By ~ 4

Author(s):

Pejman Rasti ◽

Ali Ahmad ◽

Salma Samiei ◽

Etienne Belin ◽

David Rousseau

Keyword(s):

Texture Classification ◽

Synthetic Data ◽

Real Data ◽

Local Binary Patterns ◽

High Density ◽

Weed Detection ◽

Data Set ◽

Scattering Transform ◽

Supervised Image Classification ◽

Occurrence Matrix

In this article, we assess the interest of the recently introduced multiscale scattering transform for texture classification applied for the first time in plant science. Scattering transform is shown to outperform monoscale approaches (gray-level co-occurrence matrix, local binary patterns) but also multiscale approaches (wavelet decomposition) which do not include combinatory steps. The regime in which scatter transform also outperforms a standard CNN architecture in terms of data-set size is evaluated ( 10 4 instances). An approach on how to optimally design the scatter transform based on energy contrast is provided. This is illustrated on the hard and open problem of weed detection in culture crops of high density from the top view in intensity images. An annotated synthetic data-set available under the form of a data challenge and a simulator are proposed for reproducible science (https://uabox.univ-angers.fr/index.php/s/iuj0knyzOUgsUV9). Scatter transform only trained on synthetic data shows an accuracy of 85 % when tested on real data.

Download Full-text

Vision-Based Multirotor Following Using Synthetic Learning Techniques

Sensors ◽

10.3390/s19214794 ◽

2019 ◽

Vol 19 (21) ◽

pp. 4794

Author(s):

Alejandro Rodriguez-Ramos ◽

Adrian Alvarez-Fernandez ◽

Hriday Bavle ◽

Pascual Campoy ◽

Jonathan P. How

Keyword(s):

Reinforcement Learning ◽

Object Detection ◽

Learning Strategies ◽

Motion Control ◽

Domain Adaptation ◽

Control Strategies ◽

Synthetic Data ◽

Real Data ◽

Stable Convergence ◽

Learning Techniques

Deep- and reinforcement-learning techniques have increasingly required large sets of real data to achieve stable convergence and generalization, in the context of image-recognition, object-detection or motion-control strategies. On this subject, the research community lacks robust approaches to overcome unavailable real-world extensive data by means of realistic synthetic-information and domain-adaptation techniques. In this work, synthetic-learning strategies have been used for the vision-based autonomous following of a noncooperative multirotor. The complete maneuver was learned with synthetic images and high-dimensional low-level continuous robot states, with deep- and reinforcement-learning techniques for object detection and motion control, respectively. A novel motion-control strategy for object following is introduced where the camera gimbal movement is coupled with the multirotor motion during the multirotor following. Results confirm that our present framework can be used to deploy a vision-based task in real flight using synthetic data. It was extensively validated in both simulated and real-flight scenarios, providing proper results (following a multirotor up to 1.3 m/s in simulation and 0.3 m/s in real flights).

Download Full-text

Migration moveout analysis and depth focusing

Geophysics ◽

10.1190/1.1443354 ◽

1993 ◽

Vol 58 (1) ◽

pp. 91-100 ◽

Cited By ~ 52

Author(s):

Claude F. Lafond ◽

Alan R. Levander

Keyword(s):

Heterogeneous Media ◽

A Priori ◽

Complex Structure ◽

Tomographic Reconstruction ◽

Synthetic Data ◽

Real Data ◽

Velocity Model ◽

Velocity Analysis ◽

Data Set ◽

Velocity Models

Prestack depth migration still suffers from the problems associated with building appropriate velocity models. The two main after‐migration, before‐stack velocity analysis techniques currently used, depth focusing and residual moveout correction, have found good use in many applications but have also shown their limitations in the case of very complex structures. To address this issue, we have extended the residual moveout analysis technique to the general case of heterogeneous velocity fields and steep dips, while keeping the algorithm robust enough to be of practical use on real data. Our method is not based on analytic expressions for the moveouts and requires no a priori knowledge of the model, but instead uses geometrical ray tracing in heterogeneous media, layer‐stripping migration, and local wavefront analysis to compute residual velocity corrections. These corrections are back projected into the velocity model along raypaths in a way that is similar to tomographic reconstruction. While this approach is more general than existing migration velocity analysis implementations, it is also much more computer intensive and is best used locally around a particularly complex structure. We demonstrate the technique using synthetic data from a model with strong velocity gradients and then apply it to a marine data set to improve the positioning of a major fault.

Download Full-text

Multidimensional scaling for the evaluation of a geostatistical seismic elastic inversion methodology

Geophysics ◽

10.1190/geo2013-0037.1 ◽

2014 ◽

Vol 79 (1) ◽

pp. M1-M10 ◽

Cited By ~ 22

Author(s):

Leonardo Azevedo ◽

Ruben Nunes ◽

Pedro Correia ◽

Amílcar Soares ◽

Luis Guerreiro ◽

...

Keyword(s):

Multidimensional Scaling ◽

Metric Space ◽

Synthetic Data ◽

Real Data ◽

Seismic Inversion ◽

Inversion Algorithm ◽

Data Set ◽

The Real ◽

Geostatistical Inversion ◽

Impedance Models

Due to the nature of seismic inversion problems, there are multiple possible solutions that can equally fit the observed seismic data while diverging from the real subsurface model. Consequently, it is important to assess how inverse-impedance models are converging toward the real subsurface model. For this purpose, we evaluated a new methodology to combine the multidimensional scaling (MDS) technique with an iterative geostatistical elastic seismic inversion algorithm. The geostatistical inversion algorithm inverted partial angle stacks directly for acoustic and elastic impedance (AI and EI) models. It was based on a genetic algorithm in which the model perturbation at each iteration was performed recurring to stochastic sequential simulation. To assess the reliability and convergence of the inverted models at each step, the simulated models can be projected in a metric space computed by MDS. This projection allowed distinguishing similar from variable models and assessing the convergence of inverted models toward the real impedance ones. The geostatistical inversion results of a synthetic data set, in which the real AI and EI models are known, were plotted in this metric space along with the known impedance models. We applied the same principle to a real data set using a cross-validation technique. These examples revealed that the MDS is a valuable tool to evaluate the convergence of the inverse methodology and the impedance model variability among each iteration of the inversion process. Particularly for the geostatistical inversion algorithm we evaluated, it retrieves reliable impedance models while still producing a set of simulated models with considerable variability.

Download Full-text

Seismic‐controlled nonlinear extrapolation of well parameters using neural networks

Geophysics ◽

10.1190/1.1444496 ◽

1998 ◽

Vol 63 (6) ◽

pp. 2035-2041 ◽

Cited By ~ 50

Author(s):

Zhengping Liu ◽

Jiaqi Liu

Keyword(s):

Neural Networks ◽

Joint Inversion ◽

Synthetic Data ◽

Real Data ◽

Wide Band ◽

Seismic Section ◽

Data Set ◽

Multiple Parameters ◽

Sonic Log ◽

Well Data

We present a data‐driven method of joint inversion of well‐log and seismic data, based on the power of adaptive mapping of artificial neural networks (ANNs). We use the ANN technique to find and approximate the inversion operator guided by the data set consisting of well data and seismic recordings near the wells. Then we directly map seismic recordings to well parameters, trace by trace, to extrapolate the wide‐band profiles of these parameters using the approximation operator. Compared to traditional inversions, which are based on a few prior theoretical operators, our inversion is novel because (1) it inverts for multiple parameters and (2) it is nonlinear with a high degree of complexity. We first test our algorithm with synthetic data and analyze its sensitivity and robustness. We then invert real data to obtain two extrapolation profiles of sonic log (DT) and shale content (SH), the latter a unique parameter of the inversion and significant for the detailed evaluation of stratigraphic traps. The high‐frequency components of the two profiles are significantly richer than those of the original seismic section.

Download Full-text

Is Synthetic The New Real? Performance Analysis of Time Series Generation Techniques with Focus on Network Load Forecasting

10.36227/techrxiv.17296235.v1 ◽

2021 ◽

Author(s):

Muhammad Haris Naveed ◽

Umair Hashmi ◽

Nayab Tajved ◽

Neha Sultan ◽

Ali Imran

Keyword(s):

Synthetic Data ◽

Real Data ◽

Generative Adversarial Networks ◽

Data Set ◽

Qualitative And Quantitative ◽

Adversarial Networks ◽

Network Load ◽

Quantitative Metrics ◽

Real Performance ◽

Analysis Of Time Series

This paper explores whether Generative Adversarial Networks (GANs) can produce realistic network load data that can be utilized to train machine learning models in lieu of real data. In this regard, we evaluate the performance of three recent GAN architectures on the Telecom Italia data set across a set of qualitative and quantitative metrics. Our results show that GAN generated synthetic data is indeed similar to real data and forecasting models trained on this data achieve similar performance to those trained on real data.

Download Full-text

3dinfogan: 3d Models’ Reconstruction In Infogans

Asia-Pacific Journal of Information Technology and Multimedia ◽

10.17576/apjitm-2021-1002-07 ◽

2021 ◽

Vol 10 (02) ◽

pp. 95-109

Author(s):

Du Chunqi ◽

Shinobu Hasegawa

Keyword(s):

Computer Vision ◽

Deep Learning ◽

Latent Variables ◽

Data Augmentation ◽

Synthetic Data ◽

Real Data ◽

3D Models ◽

High Quality ◽

Shape Constraint ◽

Real Objects

In computer vision and computer graphics, 3D reconstruction is the process of capturing real objects’ shapes and appearances. 3D models always can be constructed by active methods which use high-quality scanner equipment, or passive methods that learn from the dataset. However, both of these two methods only aimed to construct the 3D models, without showing what element affects the generation of 3D models. Therefore, the goal of this research is to apply deep learning to automatically generating 3D models, and finding the latent variables which affect the reconstructing process. The existing research GANs can be trained in little data with two networks called Generator and Discriminator, respectively. Generator can produce synthetic data, and Discriminator can discriminate between the generator’s output and real data. The existing research shows that InFoGAN can maximize the mutual information between latent variables and observation. In our approach, we will generate the 3D models based on InFoGAN and design two constraints, shape-constraint and parameters-constraint, respectively. Shape-constraint utilizes the data augmentation method to limit the synthetic data generated in the models’ profiles. At the same time, we also try to employ parameters-constraint to find the 3D models’ relationship corresponding to the latent variables. Furthermore, our approach will be a challenge in the architecture of generating 3D models built on InFoGAN. Finally, in the process of generation, we might discover the contribution of the latent variables influencing the 3D models to the whole network.

Download Full-text