Self-Adaptive Feature Transformation Networks for Object Detection in low luminance Images

Despite the recent improvement of object detection techniques, many of them fail to detect objects in low-luminance images. The blurry and dimmed nature of low-luminance images results in the extraction of vague features and failure to detect objects. In addition, many existing object detection methods are based on models trained on both sufficient- and low-luminance images, which also negatively affect the feature extraction process and detection results. In this article, we propose a framework called Self-adaptive Feature Transformation Network (SFT-Net) to effectively detect objects in low-luminance conditions. The proposed SFT-Net consists of the following three modules: (1) feature transformation module, (2) self-adaptive module, and (3) object detection module. The purpose of the feature transformation module is to enhance the extracted feature through unsupervisely learning a feature domain projection procedure. The self-adaptive module is utilized as a probabilistic module producing appropriate features either from the transformed or the original features to further boost the performance and generalization ability of the proposed framework. Finally, the object detection module is designed to accurately detect objects in both low- and sufficient- luminance images by using the appropriate features produced by the self-adaptive module. The experimental results demonstrate that the proposed SFT-Net framework significantly outperforms the state-of-the-art object detection techniques, achieving an average precision (AP) of up to 6.35 and 11.89 higher on the sufficient- and low- luminance domain, respectively.

Machine-learning and Automatically Segmented Retinal Biomarkers Generate Spatial Heatmaps Predictive for Standard and Low Luminance Visual Acuity in Geographic Atrophy

Objective: Predict visual function with machine-learning applied to automatically acquired quantitative imaging biomarkers in geographic atrophyDesign: Post-hoc analysis of data from a clinical trial and routine clinical care.Methods: Automated segmentation of OCT scans from 476 eyes (325 patients) with geographic atrophy. Machine learning modelling of resultant quantitative OCT (qOCT) biomarkers to predict cross-sectional visual acuity under both standard luminance (VA) and low luminance (LLVA) conditions.Main Outcome Measure: Correlation coefficient (R2) and mean absolute error (MAE) for cross-sectional VA and LLVA in Early Treatment Diabetic Retinopathy Study (ETDRS) letters.Results: Best-corrected VA under both standard luminance (R2 0.46 MAE 10.2 ETDRS letters) and low-luminance conditions (R2 0.25 MAE 12.1) could be predicted. The foveal region contributed the most (46.5%) toward model performance, with retinal pigment epithelium loss and outer retinal atrophy contributing the most (31.1%). For LLVA, however, features in the non-foveal regions were most important (74.5%), led by photoreceptor degeneration (38.9%).Conclusions: Our method of automatic qOCT segmentation demonstrates functional significance for vision in geographic atrophy, including LLVA. LLVA is itself predictive of geographic atrophy progression, implying that the predictive qOCT biomarkers provided by our model are also prognostic.

Modelling of Luminous Flux Directed to the Upper Hemisphere from Electrical Substation before and after the Refurbishment of Lighting Systems

This article deals with options for how to express the luminous flux from outdoor electrical substations of the electric transmission system. Processing of the models of light emissions before and after refurbishment of lighting systems was motivated by setting out rules for the design and erection of refurbished lighting systems in outdoor electrical substations, which are most commonly built in inhabited rural areas with low luminance of the background. The proposed model and calculations are based on requirements of international standards and recommendations for lighting of outdoor workplaces as well as on internal regulations of the ČEPS (Czech Transmission System Operator). These requirements are implemented in real electrical substations and lighting models that are extended by the calculation space of the software goniophotometer. The software goniophotometer was used to evaluate light distribution characteristics of entire electric stations in various situations. This article assesses the impact of different lighting systems installed at electrical substations from the perspective of both direct and total luminous flux directed to the upper hemisphere. It takes into account three outdoor electrical substations (420 kV) of a transmission network and their lighting systems before and after refurbishment. The aim of this article is to determine to what extent the electrical substations contribute to emissions of luminous flux to the upper hemisphere. Results from calculations on models show reduced radiation to the upper hemisphere up to 52.3% after refurbishment of obsolete lighting systems, although total installed flux actually increased due to a change in the ratio of direct and reflected luminous flux after refurbishment of lighting systems.

Nearby contours abolish the binocular advantage

That binocular viewing confers an advantage over monocular viewing for detecting isolated low luminance or low contrast objects, has been known for well over a century; however, the processes involved in combining the images from the two eyes are still not fully understood. Importantly, in natural vision, objects are rarely isolated but appear in context. It is well known that nearby contours can either facilitate or suppress detection, depending on their distance from the target and the global configuration. Here we report that at close distances collinear (but not orthogonal) flanking contours suppress detection more under binocular compared to monocular viewing, thus completely abolishing the binocular advantage, both at threshold and suprathreshold levels. In contrast, more distant flankers facilitate both monocular and binocular detection, preserving a binocular advantage up to about four times the detection threshold. Our results for monocular and binocular viewing, for threshold contrast discrimination without nearby flankers, can be explained by a gain control model with uncertainty and internal multiplicative noise adding additional constraints on detection. However, in context with nearby flankers, both contrast detection threshold and suprathreshold contrast appearance matching require the addition of both target-to-target and flank-to-target interactions occurring before the site of binocular combination. To test an alternative model, in which the interactions occur after the site of binocular combination, we performed a dichoptic contrast matching experiment, with the target presented to one eye, and the flanks to the other eye. The two models make very different predictions for abutting flanks under dichoptic conditions. Interactions after the combination site predict that the perceived contrast of the flanked target will be strongly suppressed, while interactions before the site predict the perceived contrast will be more or less veridical. The data are consistent with the latter model, strongly suggesting that the interactions take place before the site of binocular combination.

Different latencies of edge detection cause the fluttering-heart illusion

We studied the mechanism causing the fluttering-heart illusion in which the motion of an inner figure appears unsynchronized compared with that of the outer figure surrounding it although the motion of both figures is objectively synchronized in reality. Experiment 1 examined the effect of edges’ luminance contrasts. The illusion was measured under conditions where the luminance contrasts of the outer and inner figures’ edges were varied. The results indicated that the illusion occurred when the outer figure’s edge had a high luminance contrast and the inner figure’s edge had a low luminance contrast and that the illusion was reversed when the outer figure’s edge had a low luminance contrast and the inner figure’s edge had a high luminance contrast. Experiment 2 examined the effect of the first- and second-order edges. The illusion was measured under conditions where the first- and second-order edges coexisted or only the first-order edges existed. The results indicated that the illusion occurred when the outer figure had the first-order edge and the inner figure had the second-order edge, and that the illusion was reversed when the outer figure had the second-order edge and the inner figure had the first-order edge. These findings supported the hypothesis that the different latencies of edge detection cause the fluttering-heart illusion.

Spatiotemporal evolution of COVID-19 infection and detection within night light networks: comparative analysis of USA and China

The spatial distribution of population affects disease transmission, especially when shelter in place orders restrict mobility for a large fraction of the population. The spatial network structure of settlements therefore imposes a fundamental constraint on the spatial distribution of the population through which a communicable disease can spread. In this analysis we use the spatial network structure of lighted development as a proxy for the distribution of ambient population to compare the spatiotemporal evolution of COVID-19 confirmed cases in the USA and China. The Visible Infrared Imaging Radiometer Suite (VIIRS) Day/Night Band sensor on the NASA/NOAA Suomi satellite has been imaging night light at ~ 700 m resolution globally since 2012. Comparisons with sub-kilometer resolution census observations in different countries across different levels of development indicate that night light luminance scales with population density over ~ 3 orders of magnitude. However, VIIRS’ constant ~ 700 m resolution can provide a more detailed representation of population distribution in peri-urban and rural areas where aggregated census blocks lack comparable spatial detail. By varying the low luminance threshold of VIIRS-derived night light, we depict spatial networks of lighted development of varying degrees of connectivity within which populations are distributed. The resulting size distributions of spatial network components (connected clusters of nodes) vary with degree of connectivity, but maintain consistent scaling over a wide range (5 × to 10 × in area & number) of network sizes. At continental scales, spatial network rank-size distributions obtained from VIIRS night light brightness are well-described by power laws with exponents near −2 (slopes near −1) for a wide range of low luminance thresholds. The largest components (104 to 105 km2) represent spatially contiguous agglomerations of urban, suburban and periurban development, while the smallest components represent isolated rural settlements. Projecting county and city-level numbers of confirmed cases of COVID-19 for the USA and China (respectively) onto the corresponding spatial networks of lighted development allows the spatiotemporal evolution of the epidemic (infection and detection) to be quantified as propagation within networks of varying connectivity. Results for China show rapid nucleation and diffusion in January 2020 followed by rapid decreases in new cases in February. While most of the largest cities in China showed new confirmed cases approaching zero before the end of February, most of these cities also showed distinct second waves of cases in March or April. Whereas new cases in Wuhan did not approach zero until mid-March, as of December 2020 it has not yet experienced a second wave of cases. In contrast, the results for the USA show a wide range of trajectories, with an abrupt transition from slow increases in confirmed cases in a small number of network components in January and February, to rapid geographic dispersion to a larger number of components shortly before mobility reductions occurred in March. Results indicate that while most of the upper tail of the network had been exposed by the end of March, the lower tail of the component size distribution has only shown steep increases since mid-June.