Fully Convolutional Neural Network Model Research Articles

Exploring the Impact of Noise and Image Quality on Deep Learning Performance in DXA Images.

Segmentation of the femur in Dual-Energy X-ray (DXA) images poses challenges due to reduced contrast, noise, bone shape variations, and inconsistent X-ray beam penetration. In this study, we investigate the relationship between noise and certain deep learning (DL) techniques for semantic segmentation of the femur to enhance segmentation and bone mineral density (BMD) accuracy by incorporating noise reduction methods into DL models. Convolutional neural network (CNN)-based models were employed to segment femurs in DXA images and evaluate the effects of noise reduction filters on segmentation accuracy and their effect on BMD calculation. Various noise reduction techniques were integrated into DL-based models to enhance image quality before training. We assessed the performance of the fully convolutional neural network (FCNN) in comparison to noise reduction algorithms and manual segmentation methods. Our study demonstrated that the FCNN outperformed noise reduction algorithms in enhancing segmentation accuracy and enabling precise calculation of BMD. The FCNN-based segmentation approach achieved a segmentation accuracy of 98.84% and a correlation coefficient of 0.9928 for BMD measurements, indicating its effectiveness in the clinical diagnosis of osteoporosis. In conclusion, integrating noise reduction techniques into DL-based models significantly improves femur segmentation accuracy in DXA images. The FCNN model, in particular, shows promising results in enhancing BMD calculation and clinical diagnosis of osteoporosis. These findings highlight the potential of DL techniques in addressing segmentation challenges and improving diagnostic accuracy in medical imaging.

Segmentation and deep learning to digitalize the kinematics of flow-type landslides

AbstractFlow-type landslides, including subaerial and submarine debris flows, have poor spatiotemporal predictability. Therefore, researchers rely heavily on experimental evidence in revealing complex flow mechanisms and evaluating theoretical models. To measure the velocity field of experimental flows, conventional image analysis tools for measuring soil deformation and hydraulics have been borrowed. However, these tools were not developed for capturing the kinematics of fast-moving soil–water mixtures over complex terrain under non-uniform lighting conditions. In this study, a new framework based on deep learning was used to automatically digitalize the kinematics of experimental flow-type landslides. Captured images were broken into sequences and binarized using a fully convolutional neural network (FCNN). The proposed framework was demonstrated to outperform classic image processing algorithms (e.g., particle image velocimetry, trainable Weka segmentation, and thresholding algorithms) over a wide range of experimental conditions. The FCNN model was even able to process images from consumer-grade cameras under complex shadow, light, and boundary conditions. This feature is most useful for field-scale experimentation. With fewer than 15 annotated training images, the FCNN digitalized experimental flows with an accuracy of 97% in semantic segmentation.

End-to-End One-Shot Path-Planning Algorithm for an Autonomous Vehicle Based on a Convolutional Neural Network Considering Traversability Cost.

Path planning plays an important role in navigation and motion planning for robotics and automated driving applications. Most existing methods use iterative frameworks to calculate and plan the optimal path from the starting point to the endpoint. Iterative planning algorithms can be slow on large maps or long paths. This work introduces an end-to-end path-planning algorithm based on a fully convolutional neural network (FCNN) for grid maps with the concept of the traversability cost, and this trains a general path-planning model for 10 × 10 to 80 × 80 square and rectangular maps. The algorithm outputs the lowest-cost path while considering the cost and the shortest path without considering the cost. The FCNN model analyzes the grid map information and outputs two probability maps, which show the probability of each point in the lowest-cost path and the shortest path. Based on the probability maps, the actual optimal path is reconstructed by using the highest probability method. The proposed method has superior speed advantages over traditional algorithms. On test maps of different sizes and shapes, for the lowest-cost path and the shortest path, the average optimal rates were 72.7% and 78.2%, the average success rates were 95.1% and 92.5%, and the average length rates were 1.04 and 1.03, respectively.

Prediction of the proximate analysis parameters of refuse-derived fuel based on deep learning approach.

Determination of proximate characteristics can be achieved using conventional analyses methods that require a certain amount of time. In cement factories, refuse-derived fuel (RDF) is continuously fed to a kiln by a conveyor belt, so even if an inappropriate proximate characteristic is determined, it would be too late to prevent the feeding of RDF to the kiln. To overcome this problem, there is a need for instant measurement of the proximate characteristics (moisture, volatile matter, ash) that enables the feeding to be stopped. In such cases, the deep learning (DL) is a useful method based on the prediction of proximate characteristics. Therefore, in this study, the aim is to estimate the mentioned parameters developed by near-infrared spectroscopy (NIR) combined with deep learning models. For this purpose, the spectrographic measurements taken from RDF samples with an NIR spectrometer, and the results of proximate analysis in a laboratory, were used together as a dataset. A fully convolutional neural network (FCNN) and ResNet were used as a network, and they were trained using images of RDF samples and proximate analysis values. The FCNN model was more successful in prediction studies. According to the FCNN model, the results show that the models in the study can predict the moisture, ash, and volatile matter content of RDF with satisfactory R2 values between 0.979, 0.983, and 0.952.

Research on Wear Detection of End Milling Cutter Edge Based on Image Stitching

Obtaining the complete wear state of the milling cutter during processing can help predict tool life and avoid the impact of tool breakage. A cylindrical model of tool collection is proposed, which uses the collected partial pictures of the side edge to construct a panoramic picture of tool wear. After evaluating the splicing accuracy, the fully convolutional neural network (FCN) segmentation algorithm of the VGG16 structure is used to segment the panorama of the side edge of the end mill after splicing. The FCN model is built using Tensorflow to complete the image segmentation training and testing of the side edge wear area. Experimental results show that the FCN model can segment the side wear image and effectively solve the illumination change problem and different tool wear differences. Compared with the Otsu threshold adaptive segmentation algorithm and K-means clustering algorithm, the error of the extracted wear value is 1.34% to 8.93%, and the average error rate is 5.23%. This method can obtain a more intuitive panorama of the cutter side edge wear of the end milling and provide technical support for improving tool utilization rate, machining quality, and tool selection and optimization.

Toward On-Device AI and Blockchain for 6G-Enabled Agricultural Supply Chain Management

6G envisions artificial intelligence (AI) powered solutions for enhancing the quality-of-service (QoS) in the network and to ensure optimal utilization of resources. In this work, we propose an architecture based on the combination of unmanned aerial vehicles (UAVs), AI and blockchain for agricultural supply-chain management with the purpose of ensuring traceability, transparency, tracking inventories and contracts. We propose a solution to facilitate on-device AI by generating a roadmap of models with various resource-accuracy trade-offs. A fully convolutional neural network (FCN) model is used for biomass estimation through images captured by the UAV. Instead of a single compressed FCN model for deployment on UAV, we motivate the idea of iterative pruning to provide multiple task-specific models with various complexities and accuracy. To alleviate the impact of flight failure in a 6G enabled dynamic UAV network, the proposed model selection strategy will assist UAVs to update the model based on the runtime resource requirements.

Convolutional-network models to predict wall-bounded turbulence from wall quantities

Two models based on convolutional neural networks are trained to predict the two-dimensional instantaneous velocity-fluctuation fields at different wall-normal locations in a turbulent open-channel flow, using the wall-shear-stress components and the wall pressure as inputs. The first model is a fully convolutional neural network (FCN) which directly predicts the fluctuations, while the second one reconstructs the flow fields using a linear combination of orthonormal basis functions, obtained through proper orthogonal decomposition (POD), and is hence named FCN-POD. Both models are trained using data from direct numerical simulations at friction Reynolds numbers $Re_{\tau } = 180$ and 550. Being able to predict the nonlinear interactions in the flow, both models show better predictions than the extended proper orthogonal decomposition (EPOD), which establishes a linear relation between the input and output fields. The performance of the models is compared based on predictions of the instantaneous fluctuation fields, turbulence statistics and power-spectral densities. FCN exhibits the best predictions closer to the wall, whereas FCN-POD provides better predictions at larger wall-normal distances. We also assessed the feasibility of transfer learning for the FCN model, using the model parameters learned from the $Re_{\tau }=180$ dataset to initialize those of the model that is trained on the $Re_{\tau }=550$ dataset. After training the initialized model at the new $Re_{\tau }$ , our results indicate the possibility of matching the reference-model performance up to $y^{+}=50$ , with $50\,\%$ and $25\,\%$ of the original training data. We expect that these non-intrusive sensing models will play an important role in applications related to closed-loop control of wall-bounded turbulence.

Deep Learning for Effective Refugee Tent Extraction Near Syria–Jordan Border

Rukban is a desert area crossing the border between Syria and Jordan, and thousands of Syrian refugees fled into this area since the Syrian civil war in 2014. In the past few years, the number of refugee shelters for the forcibly displaced Syrian refugees in this area has increased rapidly. Estimating the location and number of refugee tents has become a key factor to maintain the sustainability of the refugee shelter camps. Manually counting the shelters is labor-intensive and sometimes prohibitive given the large quantities. In addition, these shelters/tents are usually small in size, irregular in shape, and sparsely distributed in a very large area and could be easily missed by the traditional image-analysis techniques, making the image-based approaches also challenging. In this letter, we proposed a deep fully convolutional neural network (FCN) model to extract automatically the refugee shelters/tents in the worldview-2 (WV-2) satellite images. In addition, we transferred knowledge in the pretrained VGG-16 model to improve the detection accuracy and network training convergence. We compared the proposed approach with the traditional spectral angle mapper (SAM) method, deep convolutional neural network (CNN) models, and the mask Region-based CNN (R-CNN) model. The experimental results show that the FCN model improved the overall accuracy by 4.49%, 3.54%, and 0.88% compared with the CNNs, SAM, and mask R-CNN models, and improved the precision by 34.61%, 41.99%, and 11.87%, respectively.

Exact Tile-Based Segmentation Inference for Images Larger than GPU Memory.

We address the problem of performing exact (tiling-error free) out-of-core semantic segmentation inference of arbitrarily large images using fully convolutional neural networks (FCN). FCN models have the property that once a model is trained, it can be applied on arbitrarily sized images, although it is still constrained by the available GPU memory. This work is motivated by overcoming the GPU memory size constraint without numerically impacting the final result. Our approach is to select a tile size that will fit into GPU memory with a halo border of half the network receptive field. Next, stride across the image by that tile size without the halo. The input tile halos will overlap, while the output tiles join exactly at the seams. Such an approach enables inference to be performed on whole slide microscopy images, such as those generated by a slide scanner. The novelty of this work is in documenting the formulas for determining tile size and stride and then validating them on U-Net and FC-DenseNet architectures. In addition, we quantify the errors due to tiling configurations which do not satisfy the constraints, and we explore the use of architecture effective receptive fields to estimate the tiling parameters.

A deep learning approach for the quantification of lower tear meniscus height

The quantification of the tear meniscus height can be helpful in the diagnosis of Dry Eyes Disease. This paper presents a method for automatic quantitation of lower tear meniscus height (TMH) with fully convolutional neural networks (FCNN) and investigate its performance and efficacy compared to manual measurements. A total of 485 images from 217 subjects were acquired with a mainstream corneal topographer and then divided these images into the development and testing set respectively. The development set was used to train the FCNN models, while the testing set to evaluate the performance of the models. TMH of each image was assessed by the proposed method based on the corresponding segmentation mask of tear meniscus and compared against the manual results. The tear meniscus of each image in the testing set was segmented by the FCNN. Five-fold cross validation revealed an overall average intersection of Union (IoU) of 82.5 %, a F1-score of 90.1 % for tear meniscus segmentation. The algorithm results of TMH had a higher correlation (r = 0.965, p < 0.001) with the ground-truth compared with the manual obtained results (r = 0.898, p < 0.001). The curve of TMH was plotted to reveal the spatial variation along the lower eyelid from nasal and temporal. Higher TMH were found at nasal (median: 0.26 mm) and temporal (0.27 mm) canthi compared with the locations right under the pupil center (0.19 mm). On the experimental data, the proposed method provided reliable TMH results with a higher consistency and efficacy. It was expected to form an assistive tool in TMH quantitation and subsequently and screening on Dry Eyes Disease.

Abstract PO-014: Deep learning-based segmentation accurately captures histological features in cancer-free lymph nodes of breast cancer patients

Abstract Background Histological changes in the cancer-free lymph nodes (LNs) can add to the risk prediction of developing distant metastasis for LN-positive breast cancer patients. We have previously shown that the formation of germinal centres and enlarged sinus are premetastatic features and indicative of superior prognosis. As Whole Slide Images (WSI) demonstrated suitability for detection of LN metastasis with high accuracy based on Convolutional Neural Networks (CNN), we extended the CNN assessments to the detection and segmentation of these premetastatic features. Materials For 136 breast cancer patients, retrospectively collected from Guy’s Hospital, 1,422 Whole Slide Images (WSI) of axillary LNs (containing 1 to 3 LNs per WSI) and primary invasive tumours were 40x scanned by the Hamamatsu Nanozoomer. Two pathologists manually annotated germinal centres and sinuses in 111 of these images facilitating the ground-truth masks for a binary segmentation task using deep learning models. Results Each WSI was split into a series of overlapping patches along with corresponding ground-truth masks at 2.5x, 5x and 10x. Fully Convolutional Neural Networks (FCNs) based on the U-Net architecture were trained on patches from 95 annotated WSI to perform a segmentation for both germinal centres and sinus, with the remaining images used for validation (n= 6) and test sets (n= 10). In addition to a vanilla U-Net encoder-decoder architecture, a model with an attention mechanism to filter feature maps along skip connections was also explored. Additionally, a U-Net with dilated convolutions at each convolution block was used in a multi-scale mechanism that combines features from three different resolutions. To increase the robustness of the FCNs, training data was augmented using random image rotation and different colour perturbation transformations. Predictions were made on regions of WSIs containing whole LNs and were compared with the pathologist annotations on holdout testing images. The Dice coefficient was used to compare the predicted segmentations with the ground-truth annotations. Standard U-Net and attention U-Net architectures achieved an average Dice coefficient of 0.75 across the holdout testing images at 2.5x magnification for germinal centres, with individual images achieving a Dice score above 0.8. Multiscale models achieved the best results at 10x magnification with an average Dice score of 0.85 for germinal centres and 0.75 for sinus. The trained models were then applied to &amp;gt; 1,300 unseen and non-annotated WSIs of axillary LNs. Conclusion Our developed FCN models allow an automated, precise, reproducible and time-efficient approach to annotate germinal centres and sinuses in cancer-free and metastatic LNs. Trained models are currently applied to &amp;gt;1,300 unseen and non-annotated WSIs of axillary LNs, and will demonstrate their comparable performance with manual pathological assessment. Citation Format: Gregory Verghese, Anita Grigoriadis, Amit Sethi, Amit Lohan, Nikhil Cherian, Swati Meena, Harry Chinque, Fangfang Liu, Ed Martin, Tristan Clark, Cheryl Gillet, Julie Owen, Swapnil Rane. Deep learning-based segmentation accurately captures histological features in cancer-free lymph nodes of breast cancer patients [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr PO-014.

Lane Line Detection via Deep Learning Based- Approach Applying Two Types of Input into Network Model

&#x0D; &#x0D; &#x0D; Lane line detection is one of the important modules for Advanced Driver-Assistance System (ADAS) that are applied in the autonomous vehicle. This module work by exhibit the position of the road lane marking and providing the details of the geometrical features of the lane line structures into the intelligent system. This paper proposes the lane line marking detection using Fully Convolutional Neural Network (FCN) model by investigating the two types of input fed into the networks. RGB- channel (Red, Green, Blue) and Canny edge were used as the inputs to develop in the FCN model. The FCN approach has been proposed as one of the solution methods in mitigating the road lane detection issues due to its great performance in the application of objects detection in image or video. Previously, the RGB-channel is widely applied in the deep learning method meanwhile, the Canny-edge input has not been applied yet in the deep learning method. Therefore, this study investigates the further performance of this model by applying the canny edge as addition input besides applying only the RGB-channel. The data collections were acquired from real-time data collection. The result shows that the FCN model with the canny edge achieved a slight improvement with 96% compared to FCN with the RGB-channel with 92%.&#x0D; &#x0D; &#x0D;

Prediction of wall-bounded turbulence from wall quantities using convolutional neural networks

A fully-convolutional neural-network model is used to predict the streamwise velocity fields at several wall-normal locations by taking as input the streamwise and spanwise wall-shear-stress planes in a turbulent open channel flow. The training data are generated by performing a direct numerical simulation (DNS) at a friction Reynolds number of Reτ = 180. Various networks are trained for predictions at three inner-scaled locations (y+ = 15, 30, 50) and for different time steps between input samples Δt+ s. The inherent non-linearity of the neural-network model enables a better prediction capability than linear methods, with a lower error in both the instantaneous flow fields and turbulent statistics. Using a dataset with higher Δt+ s improves the generalization at all the considered wall-normal locations, as long as the network capacity is sufficient to generalize over the dataset. The use of a multiple-output network, with parallel dedicated branches for two wall-normal locations, does not provide any improvement over two separated single-output networks, other than a moderate saving in training time. Training time can be effectively reduced, by a factor of 4, via a transfer learning method that initializes the network parameters using the optimized parameters of a previously-trained network.

Measurement of Conditional Relatedness Between Genes Using Fully Convolutional Neural Network.

Measuring conditional relatedness, the degree of relation between a pair of genes in a certain condition, is a basic but difficult task in bioinformatics, as traditional co-expression analysis methods rely on co-expression similarities, well known with high false positive rate. Complement with prior-knowledge similarities is a feasible way to tackle the problem. However, classical combination machine learning algorithms fail in detection and application of the complex mapping relations between similarities and conditional relatedness, so a powerful predictive model will have enormous benefit for measuring this kind of complex mapping relations. To this need, we propose a novel deep learning model of convolutional neural network with a fully connected first layer, named fully convolutional neural network (FCNN), to measure conditional relatedness between genes using both co-expression and prior-knowledge similarities. The results on validation and test datasets show FCNN model yields an average 3.0% and 2.7% higher accuracy values for identifying gene–gene interactions collected from the COXPRESdb, KEGG, and TRRUST databases, and a benchmark dataset of Xiao-Yong et al. research, by grid-search 10-fold cross validation, respectively. In order to estimate the FCNN model, we conduct a further verification on the GeneFriends and DIP datasets, and the FCNN model obtains an average of 1.8% and 7.6% higher accuracy, respectively. Then the FCNN model is applied to construct cancer gene networks, and also calls more practical results than other compared models and methods. A website of the FCNN model and relevant datasets can be accessed from https://bmbl.bmi.osumc.edu/FCNN.

Artificial intelligence in neuropathology: deep learning-based assessment of tauopathy

Accumulation of abnormal tau in neurofibrillary tangles (NFT) occurs in Alzheimer disease (AD) and a spectrum of tauopathies. These tauopathies have diverse and overlapping morphological phenotypes that obscure classification and quantitative assessments. Recently, powerful machine learning-based approaches have emerged, allowing the recognition and quantification of pathological changes from digital images. Here, we applied deep learning to the neuropathological assessment of NFT in postmortem human brain tissue to develop a classifier capable of recognizing and quantifying tau burden. The histopathological material was derived from 22 autopsy brains from patients with tauopathies. We used a custom web-based informatics platform integrated with an in-house information management system to manage whole slide images (WSI) and human expert annotations as ground truth. We utilized fully annotated regions to train a deep learning fully convolutional neural network (FCN) implemented in PyTorch against the human expert annotations. We found that the deep learning framework is capable of identifying and quantifying NFT with a range of staining intensities and diverse morphologies. With our FCN model, we achieved high precision and recall in naive WSI semantic segmentation, correctly identifying tangle objects using a SegNet model trained for 200 epochs. Our FCN is efficient and well suited for the practical application of WSIs with average processing times of 45 min per WSI per GPU, enabling reliable and reproducible large-scale detection of tangles. We measured performance on test data of 50 pre-annotated regions on eight naive WSI across various tauopathies, resulting in the recall, precision, and an F1 score of 0.92, 0.72, and 0.81, respectively. Machine learning is a useful tool for complex pathological assessment of AD and other tauopathies. Using deep learning classifiers, we have the potential to integrate cell- and region-specific annotations with clinical, genetic, and molecular data, providing unbiased data for clinicopathological correlations that will enhance our knowledge of the neurodegeneration.

Brain and lesion segmentation in multiple sclerosis using fully convolutional neural networks: A large-scale study

Objective: To investigate the performance of deep learning (DL) based on fully convolutional neural network (FCNN) in segmenting brain tissues in a large cohort of multiple sclerosis (MS) patients. Methods: We developed a FCNN model to segment brain tissues, including T2-hyperintense MS lesions. The training, validation, and testing of FCNN were based on ~1000 magnetic resonance imaging (MRI) datasets acquired on relapsing–remitting MS patients, as a part of a phase 3 randomized clinical trial. Multimodal MRI data (dual-echo, FLAIR, and T1-weighted images) served as input to the network. Expert validated segmentation was used as the target for training the FCNN. We cross-validated our results using the leave-one-center-out approach. Results: We observed a high average (95% confidence limits) Dice similarity coefficient for all the segmented tissues: 0.95 (0.92–0.98) for white matter, 0.96 (0.93–0.98) for gray matter, 0.99 (0.98–0.99) for cerebrospinal fluid, and 0.82 (0.63–1.0) for T2 lesions. High correlations between the DL segmented tissue volumes and ground truth were observed (R2 > 0.92 for all tissues). The cross validation showed consistent results across the centers for all tissues. Conclusion: The results from this large-scale study suggest that deep FCNN can automatically segment MS brain tissues, including lesions, with high accuracy.

DeepVOG: Open-source pupil segmentation and gaze estimation in neuroscience using deep learning

BackgroundA prerequisite for many eye tracking and video-oculography (VOG) methods is an accurate localization of the pupil. Several existing techniques face challenges in images with artifacts and under naturalistic low-light conditions, e.g. with highly dilated pupils. New methodFor the first time, we propose to use a fully convolutional neural network (FCNN) for segmentation of the whole pupil area, trained on 3946 VOG images hand-annotated at our institute. We integrate the FCNN into DeepVOG, along with an established method for gaze estimation from elliptical pupil contours, which we improve upon by considering our FCNN's segmentation confidence measure. ResultsThe FCNN output simultaneously enables us to perform pupil center localization, elliptical contour estimation and blink detection, all with a single network and with an assigned confidence value, at framerates above 130 Hz on commercial workstations with GPU acceleration. Pupil centre coordinates can be estimated with a median accuracy of around 1.0 pixel, and gaze estimation is accurate to within 0.5 degrees. The FCNN is able to robustly segment the pupil in a wide array of datasets that were not used for training. Comparison with existing methodsWe validate our method against gold standard eye images that were artificially rendered, as well as hand-annotated VOG data from a gold-standard clinical system (EyeSeeCam) at our institute. ConclusionsOur proposed FCNN-based pupil segmentation framework is accurate, robust and generalizes well to new VOG datasets. We provide our code and pre-trained FCNN model open-source and for free under www.github.com/pydsgz/DeepVOG.

Falls Risk Classification of Older Adults Using Deep Neural Networks and Transfer Learning.

Prior research in falls risk classification using inertial sensors has relied on the use of engineered features, which has resulted in a feature space containing hundreds of features that are likely redundant and possibly irrelevant. In this paper, we propose using fully convolutional neural networks (FCNNs) to classify older adults at low or high risk of falling using inertial sensor data collected from a smartphone. Due to the limited nature of older adult inertial gait datasets, we first pre-train the FCNN models using a publicly available dataset for pedestrian activity recognition. Then via transfer learning, we train the network for falls risk classification. We show that via transfer learning, our falls risk classifier obtains an area under the receiver operating characteristic curve of 93.3%, which is 10.6% higher than the equivalent model trained without the use of transfer learning. Additionally, we show that our method outperforms other standard machine learning classifiers trained on features developed in prior research.

Fully Convolutional Neural Network for Malaysian Road Lane Detection

Recently, a deep learning, Fully Convolutional Neural Network (FCN) has been widely studied because it can demonstrate promising results in the application of detection of objects in an image or video. Hence, the FCN approach has been proposed as one of the solution methods in mitigating the issues pertinent to Malaysia’s road lane detection. Previously, FCN model for lane detection has not been tested in Malaysian road conditions. Therefore, this study investigates the further performance of this model in the Malaysia. The network model is trained and validated using the datasets obtained from Machine Learning NanoDegree. In addition, the real-time data collection has been conducted to collect the data sets for the testing at the highway and urban areas in Malaysia. Then, the collected data is used to test the performance of the FCN network in detecting the lane markings on Malaysia road. The results demonstrated that the FCN method is achieving 99% of the training and validation accuracy.

Deeply Supervised Salient Object Detection with Short Connections.

Recent progress on salient object detection is substantial, benefiting mostly from the explosive development of Convolutional Neural Networks (CNNs). Semantic segmentation and salient object detection algorithms developed lately have been mostly based on Fully Convolutional Neural Networks (FCNs). There is still a large room for improvement over the generic FCN models that do not explicitly deal with the scale-space problem. The Holistically-Nested Edge Detector (HED) provides a skip-layer structure with deep supervision for edge and boundary detection, but the performance gain of HED on saliency detection is not obvious. In this paper, we propose a new salient object detection method by introducing short connections to the skip-layer structures within the HED architecture. Our framework takes full advantage of multi-level and multi-scale features extracted from FCNs, providing more advanced representations at each layer, a property that is critically needed to perform segment detection. Our method produces state-of-the-art results on 5 widely tested salient object detection benchmarks, with advantages in terms of efficiency (0.08 seconds per image), effectiveness, and simplicity over the existing algorithms. Beyond that, we conduct an exhaustive analysis of the role of training data on performance. We provide a training set for future research and fair comparisons.