Convolution Process Research Articles

STGSA: A Novel Spatial-Temporal Graph Synchronous Aggregation Model for Traffic Prediction

The success of intelligent transportation systems relies heavily on accurate traffic prediction, in which how to model the underlying spatial-temporal information from traffic data has come under the spotlight. Most existing frameworks typically utilize separate modules for spatial and temporal correlations modeling. However, this stepwise pattern may limit the effectiveness and efficiency in spatial-temporal feature extraction and cause the overlook of important information in some steps. Furthermore, it is lacking sufficient guidance from prior information while modeling based on a given spatial adjacency graph (e.g., deriving from the geodesic distance or approximate connectivity), and may not reflect the actual interaction between nodes. To overcome those limitations, our paper proposes a spatial-temporal graph synchronous aggregation (STGSA) model to extract the localized and long-term spatial-temporal dependencies simultaneously. Specifically, a tailored graph aggregation method in the vertex domain is designed to extract spatial and temporal features in one graph convolution process. In each STGSA block, we devise a directed temporal correlation graph to represent the localized and long-term dependencies between nodes, and the potential temporal dependence is further fine-tuned by an adaptive weighting operation. Meanwhile, we construct an elaborated spatial adjacency matrix to represent the road sensor graph by considering both physical distance and node similarity in a data-driven manner. Then, inspired by the multi-head attention mechanism which can jointly emphasize information from different representation subspaces, we construct a multi-stream module based on the STGSA blocks to capture global information. It projects the embedding input repeatedly with multiple different channels. Finally, the predicted values are generated by stacking several multi-stream modules. Extensive experiments are constructed on six real-world datasets, and numerical results show that the proposed STGSA model significantly outperforms the benchmarks.

Deep Learning Framework for the Prediction of Childhood Medulloblastoma

This research work develops new and better prognostic markers for predicting Childhood MedulloBlastoma (CMB) using a well-defined deep learning architecture. A deep learning architecture could be designed using ideas from image processing and neural networks to predict CMB using histopathological images. First, a convolution process transforms the histopathological image into deep features that uniquely describe it using different two-dimensional filters of various sizes. A 10-layer deep learning architecture is designed to extract deep features. The introduction of pooling layers in the architecture reduces the feature dimension. The extracted and dimension-reduced deep features from the arrangement of convolution layers and pooling layers are used to classify histopathological images using a neural network classifier. The performance of the CMB classification system is evaluated using 1414 (10× magnification) and 1071 (100× magnification) augmented histopathological images with five classes of CMB such as desmoplastic, nodular, large cell, classic, and normal. Experimental results show that the average classification accuracy of 99.38% (10×) and 99.07% (100×) is attained by the proposed CNB classification system.

Canny and Morphological Approaches to Calculating Area and Perimeter of Two-Dimensional Geometry

Calculating area and perimeter in real-world conditions has its challenges. The actual conditions include applications in the medical field to measure the presence of tumors or the condition of human organs and applications in geography to measure specific areas on a map; applications in architecture often calculate the area and perimeter of buildings, interior design, exterior design, and other uses. Technology can make it easier with automatic calculations. Mathematical methods and computer vision techniques are required to create automated systems. The Canny method is usually used, which is good enough for detecting edges but not sufficient for measuring irregular geometric shapes. This paper aims to calculate the area and perimeter of a geometric shape using the Canny method and geometry. Data samples in various forms are used in this study. Calculating area and perimeter using the Canny method involves obtaining the length (X,Y) of the RGB image converted to HSV. Edge detection values are used to calculate the area and perimeter of objects. The morphological method uses binary image input as input data. Then proceed to the convolution process with structuring and calculating the area and circumference of objects. Based on the research results, calculating the area and circumference of objects is more effective using morphological methods. However, the level of accuracy is affected by the selection of structuring elements (strels) which must be optimal and global.

R-ProjNet: an optimal rotated-projection neural network for wood segmentation from point clouds

ABSTRACT This work aims to provide a deep learning framework to segment woods from tree point clouds. We develop a novel preprocessing layer before the classical sampling and convolution structure called the projection layer to organize 3D point clouds into 2D points. Input data are transformed into projection data along axis and planes for the subsequent convolution process, which helps decrease the complexity of networks. In order to obtain optimal and effective projection data for capturing local features, we formulate the 2D transformation in the learning process using two learnable angle parameters. The projection map is updated in the learning process for capturing geometric structure information, which plays an important role in wood point segmentation. Experiments show that we have achieved the loss and misclassification error of 0.41% and 8%, respectively, on wood points extraction from handheld laser scanning data. Besides, we also achieve the correctness, completeness and -score of 90.4%, 91.5% and 0.91, respectively, in a public vehicle laser scanning dataset.

TSRC: A Deep Learning Model for Precipitation Short-Term Forecasting over China Using Radar Echo Data

Currently, most deep learning (DL)-based models for precipitation forecasting face two conspicuous issues: the smoothing effect in the precipitation field and the degenerate effect of forecasting precipitation intensity. Therefore, this study proposes “time series residual convolution (TSRC)”, a DL-based convolutional neural network for precipitation nowcasting over China with a lead time of 3 h. The core idea of TSRC is it compensates the current local cues with previous local cues during convolution processes, so more contextual information and less uncertain features would remain in deep networks. We use four years’ radar echo reflectivity data from 2017 to 2020 for model training and one year’s data from 2021 for model testing and compare it with two commonly used nowcasting models: optical flow model (OF) and UNet. Results show that TSRC obtains better forecasting performances than OF and UNet with a relatively high probability of detection (POD), low false alarm rate (FAR), small mean absolute error (MAE) and high structural similarity index (SSIM), especially at longer lead times. Meanwhile, the results of two case studies suggest that TSRC still introduces smoothing effects and slightly outperforms UNet at longer lead times. The most considerable result is that our model can forecast high-intensity radar echoes even for typhoon rainfall systems, suggesting that the degenerate effect of forecasting precipitation intensity can be improved by our model. Future works will focus on the combination of multi-source data and the design of the model’s architecture to gain further improvements in precipitation short-term forecasting.

Enhanced broadband Si-based optoelectronic synapse for artificial visual applications

Broadband sensibility and efficient image recognition are crucial for perception and memory storage in remote sensing, reconnaissance and environmental monitoring inspired by the human cognitive system. Although the amorphous silicon (a-Si) thin film is compatible to the complementary-metal–oxidesemiconductor (CMOS) technology, implementing in-sensor broadband recognition and memory based on intrinsic a-Si is still a challenge due to the band gap restrictions and the limited spectrum sensing ability. Herein, through doping of Ru (Ruthenium), optoelectronic devices based on the band engineered a-Si:Ru films have been fabricated with a silicon-oxide heterostructure, demonstrating comprehensive broadband image recognition, memory and convolutional processing simultaneously. Upon regulating the trapping and releasing of the photogenerated electron-hole pairs, the devices could mimic various synaptic functions such as excitatory postsynaptic current (EPSC), paired-pulse facilitation (PPF), transition from short- to long-term potentiation, lower switching energy and adjustable photocurrent analog states. Moreover, the in-sensor broadband convolutional processing has enhanced the high recognition accuracy for multi-band visuals, indicating a significant opportunity for future neuromorphic sensory applications and advanced robot visual systems.

Latent Features Embedded Dynamic Graph Evolution Deep Clustering Network

As a typical unsupervised machine learning task, clustering is always a hot research topic. Motivated by deep learning approaches, deep clustering has become prevalent in recent years, and achieves appealing performance. Most of current deep clustering methods focus on learning a new discriminative representation to enhance separability of original data, i.e., autoencoder, multilayer perceptrons and deep belief networks. However, the structure information which is very important in unsupervised learning, attracts little attention in previous deep feature representation learning clustering works. In this paper, a dynamic graph evolution based graph convolutional network (DGE-GCN) is introduced for clustering task, in which the data structural information and learned latent features are integrated into a unified network for deep clustering. Instead of using a fixed graph during the graph convolution process, we design a dynamic graph evolution strategy to refine the initial graph which could be not accurate. In addition, the latent representations learned by an autoencoder are embedded for refining the graph in a layer-wise manner. In such a manner, the latent features can help improve the graph structure, while the refined graph can in turn constrain the autoencoder to learn more discriminative features. In order to unify the graph convolutional network branch and autoencoder branch, a dual self-supervised mechanism is designed to guide the parameter learning of the whole network architecture. Comprehensive experiments demonstrate that the proposed network consistently performs better than several state-of-the-art methods on various benchmark datasets.

FPSNET: An Architecture for Neural-Network-Based Feature Point Extraction for SLAM

The hardware architecture of a deep-neural-network-based feature point extraction method is proposed for the simultaneous localization and mapping (SLAM) in robotic applications, which is named the Feature Point based SLAM Network (FPSNET). Some key techniques are deployed to improve the hardware and power efficiency. The data path is devised to reduce overall off-chip memory accesses. The intermediate data and partial sums resulting in the convolution process are stored in available on-chip memories, and optimized hardware is employed to compute the one-point activation function. Meanwhile, address generation units are used to avoid data overlapping in memories. The proposed FPSNET has been designed in 65 nm CMOS technology with a core area of 8.3 mm2. This work reduces the memory overhead by 50% compared to traditional data storage for activation and overall by 35% for on-chip memories. The synthesis and simulation results show that it achieved a 2.0× higher performance compared with the previous design while achieving a power efficiency of 1.0 TOPS/W, which is 2.4× better than previous work. Compared to other ASIC designs with close peak throughput and power efficiency performance, the presented FPSNET has the smallest chip area (at least 42.4% reduction).

FIRE: knowledge-enhanced recommendation with feature interaction and intent-aware attention networks.

To solve the information overload issue and enhance the user experience of various web applications, recommender systems aim to better model user interests and preferences. Knowledge Graphs (KGs), consisting of real-world objective facts and fruitful entities, play a vital role in recommender systems. Recently, a technological trend has been to develop end-to-end Graph Neural Networks (GNNs)-based knowledge-aware recommendation (a.k.a., Knowledge Graph Recommendation, KGR) models. Unfortunately, current GNNs-based KGR approaches focus on how to capture high-order feature information on KGs while neglecting the following two crucial limitations: 1) The explicitly high-order feature interaction and fusion mechanism and 2) The valid user intent modelling mechanism. As such, these issues lead to insufficient user/item representation learning capability and unsatisfactory KGR performance. In this work, we present a novel Knowledge-enhanced Re commendation with F eature I nteraction and Intent-aware Attention Networks (FIRE) to address the latent intent modelling and high-order feature interaction deficiencies ignored by existing KGR methods. Based on the prototype user/item representation learning leveraging the GNNs-based approach, our model offers the following major improvements: One is the innovative use of Convolutional Neural Networks (CNNs) that perform vertical convolutional (a.k.a., bit-level convolutional) and horizontal convolutional (a.k.a., vector-level convolutional) processes to model multi-granular high-order feature interactions to enhance item-side representation learning. Another is to model users' latent intent factors by utilizing a two-level attention mechanism (i.e., node- and intent-level attention mechanism) to enhance user-side representation learning. Extensive experiments on three KGs domain public datasets demonstrate that our method outperforms the existing state-of-the-art baseline. Last but not least, numerous ablation- and model studies demystify the working mechanism and elucidate the plausibility of the proposed model.

Characterization of Subsurface Hydrogeological Structures With Convolutional Conditional Neural Processes on Limited Training Data

AbstractOne of the main issues in the application of statistical‐learning‐based methods to the characterization of hydrological phenomena is the complex parameterization of the high‐quality reconstruction. The difficulty of obtaining enough training data and generating multiple‐scale structures hinder the applications of deep‐learning‐based methods in hydrogeological modeling. Hydrogeological modeling problems can be regarded as stochastic processes, and we propose a novel hydrogeological modeling approach based on convolutional conditional neural processes (GM‐ConvCNPs), a meta‐learning approach for dealing with stochastic processes. In this work, GM‐ConvCNP is used to reconstruct the entire spatial structures of subsurface hydrological attributes and channels from a limited amount of conditioning data. To achieve 3‐D characterization of subsurface structures, 3‐D GM‐ConvCNP is proposed according to the spatial distribution characteristics of 3‐D conditioning data. Compared to other deep‐learning‐based methods, we use a small amount of training data and obtain positive model generalization capabilities. Four data sets of categorical and continuous hydrogeological structures are exploited in the experiments. Various validation tests including variograms, connectivity functions, and multi‐dimensional scaling (MDS) map are used to evaluate the quality of generated realizations. The proposed approach is able to significantly reduce training consumption and improve the performance of realizations compared to a set of different benchmark tests. The experiments confirm that the GM‐ConvCNP model can extract heterogeneous patterns by using meta‐learning from limited training data and reconstruct multiple‐scale hydrogeological structures. A case study demonstrates that our method can be applied to characterize multiple‐source hydrological variables.

Short-Term and Medium-Term Electricity Sales Forecasting Method Based on Deep Spatio-Temporal Residual Network

The forecasting of electricity sales is directly related to the power generation planning of power enterprises and the progress of the generation tasks. Aiming at the problem that traditional forecasting methods cannot properly deal with the actual data offset caused by external factors, such as the weather, season, and spatial attributes, this paper proposes a method of electricity sales forecasting based on a deep spatio-temporal residual network (ST-ResNet). The method not only relies on the temporal correlation of electricity sales data but also introduces the influence of external factors and spatial correlation, which greatly enhances the fitting degree of each parameter of the model. Moreover, the residual module and the convolution module are fused to effectively reduce the damage of the deep convolutional process to the training effectiveness. Finally, the three comparison experiments of the ultra-short term, short term and medium term show that the MAPE forecasted by the ST-ResNet model is at least 2.69% lower than that of the RNN and other classical Deep Learning models, that its RMSE is at least 36.2% lower, and that its MAD is at least 34.2% lower, which is more obvious than the traditional methods. The effectiveness and feasibility of the ST-ResNet model in the short-term forecasting of electricity sales are verified.

Graph based multi-scale neighboring topology deep learning for kidney and tumor segmentation

Objective. Effective learning and modelling of spatial and semantic relations between image regions in various ranges are critical yet challenging in image segmentation tasks. Approach. We propose a novel deep graph reasoning model to learn from multi-order neighborhood topologies for volumetric image segmentation. A graph is first constructed with nodes representing image regions and graph topology to derive spatial dependencies and semantic connections across image regions. We propose a new node attribute embedding mechanism to formulate topological attributes for each image region node by performing multi-order random walks (RW) on the graph and updating neighboring topologies at different neighborhood ranges. Afterwards, multi-scale graph convolutional autoencoders are developed to extract deep multi-scale topological representations of nodes and propagate learnt knowledge along graph edges during the convolutional and optimization process. We also propose a scale-level attention module to learn the adaptive weights of topological representations at multiple scales for enhanced fusion. Finally, the enhanced topological representation and knowledge from graph reasoning are integrated with content features before feeding into the segmentation decoder. Main results. The evaluation results over public kidney and tumor CT segmentation dataset show that our model outperforms other state-of-the-art segmentation methods. Ablation studies and experiments using different convolutional neural networks backbones show the contributions of major technical innovations and generalization ability. Significance. We propose for the first time an RW-driven MCG with scale-level attention to extract semantic connections and spatial dependencies between a diverse range of regions for accurate kidney and tumor segmentation in CT volumes.

Localization of mixed intracranial hemorrhages by using a ghost convolution-based YOLO network

Intracranial hemorrhage (ICH) is a serious medical condition that must be diagnosed in a stipulated time through computed tomography (CT) imaging modality. However, the neurologist must initially confirm the specific type of hemorrhage to prescribe an effective treatment. Although conventional image processing and convolution-based deep learning models can effectively perform multiclass classification tasks, they fail to classify if a CT input image contains multiple hemorrhages in a single slice and takes a lot of time to make the final predictions. To overcome these two difficulties, we proposed a novel YOLOv5x-GCB model that can be able to detect multiple hemorrhages with limited resources by employing a ghost convolution process. The advantage of ghost convolution is that it produces the same number of feature maps as vanilla convolution while using less expensive linear operations. Another feature of the proposed model is that it uses the mosaic augmentation technique throughout the training to improve the accuracy of mixed hemorrhage detection. A brain hemorrhage extended dataset containing 21,132 slices from 205 positive patients was used in training and validating the proposed model. To test the robustness of the proposed model, we created a separate dataset with the existing segmentation data, which are available in PhysioNet. As a result, the proposed model achieved an overall precision, recall, F1- score, and mean average precision of 92.1%, 88.9%, 90%, and 93.1%, respectively. In addition to these metrics, other parameters were used in evaluating the proposed model and checking its lightweight capability in terms of memory size and computational time. Results showed that our proposed model can be used in real-time clinical diagnosis by using either embedded devices or cloud services.

Continuous Capsule Network Method for Improving Electroencephalogram-Based Emotion Recognition

The convolution process in the Capsule Network method can result in a loss of spatial data from the Electroencephalogram signal, despite its ability to characterize spatial information from Electroencephalogram signals. Therefore, this study applied the Continuous Capsule Network method to overcome problems associated with emotion recognition based on Electroencephalogram signals using the optimal architecture of the (1) 1st, 2nd, 3rd, and 4th Continuous Convolution layers with values of 64, 128, 256, and 64, respectively, and (2) kernel sizes of 2×2×4, 2×2×64, and 2×2×128 for the 1st, 2nd, and 3rd Continuous Convolution layers, and 1×1×256 for the 4th. Several methods were also used to support the Continuous Capsule Network process, such as the Differential Entropy and 3D Cube methods for the feature extraction and representation processes. These methods were chosen based on their ability to characterize spatial and low-frequency information from Electroencephalogram signals. By testing the DEAP dataset, these proposed methods achieved accuracies of 91.35, 93.67, and 92.82% for the four categories of emotions, two categories of arousal, and valence, respectively. Furthermore, on the DREAMER dataset, these proposed methods achieved accuracies of 94.23, 96.66, and 96.05% for the four categories of emotions, the two categories of arousal, and valence, respectively. Finally, on the AMIGOS dataset, these proposed methods achieved accuracies of 96.20, 97.96, and 97.32% for the four categories of emotions, the two categories of arousal, and valence, respectively. Doi: 10.28991/ESJ-2023-07-01-09 Full Text: PDF

A Fully Natural Gradient Scheme for Improving Inference of the Heterogeneous Multioutput Gaussian Process Model.

A recent novel extension of multioutput Gaussian processes (GPs) handles heterogeneous outputs, assuming that each output has its own likelihood function. It uses a vector-valued GP prior to jointly model all likelihoods' parameters as latent functions drawn from a GP with a linear model of coregionalization (LMC) covariance. By means of an inducing points' framework, the model is able to obtain tractable variational bounds amenable to stochastic variational inference (SVI). Nonetheless, the strong conditioning between the variational parameters and the hyperparameters burdens the adaptive gradient optimization methods used in the original approach. To overcome this issue, we borrow ideas from variational optimization introducing an exploratory distribution over the hyperparameters, allowing inference together with the posterior's variational parameters through a fully natural gradient (NG) optimization scheme. Furthermore, in this work, we introduce an extension of the heterogeneous multioutput model, where its latent functions are drawn from convolution processes. We show that our optimization scheme can achieve better local optima solutions with higher test performance rates than adaptive gradient methods for both the LMC and the convolution process model. We also show how to make the convolutional model scalable by means of SVI and how to optimize it through a fully NG scheme. We compare the performance of the different methods over the toy and real databases.

Correlated Chained Gaussian Processes for Modelling Citizens Mobility Using a Zero-Inflated Poisson Likelihood

Modelling the mobility of people in a city depends on counting data with inherent problems of overdispersion. Such dispersion issues are caused by massive amounts of data with zero values. Though traditional machine learning models have been used to overcome said problems, they lack the ability to appropriately model the spatio-temporal correlations in data. To improve the modelling of such spatio-temporal correlations, in this work we propose to model the citizens mobility, for the Chinese city of Guangzhou, by means of a Zero-inflated Poisson likelihood in conjunction with Gaussian process priors generated from convolution processes. We follow the idea of chaining the likelihood’s parameters to latent functions drawn from Gaussian process priors; this way allowing a higher flexibility to model heteroscedasticity. Additionally, we derive a stochastic variational inference framework that allow us to use two types of convolution process models in the context of large datasets: correlated chained Gaussian processes with a convolution processes model, and correlated chained Gaussian processes with variational inducing kernels. We present quantitative and qualitative results comparing the performance between Negative Binomial and Zero-inflated Poisson likelihoods, both in combination with three types of Gaussian process priors: a linear model of coregionalisation, and our two proposed methods based on a convolution processes model and variational inducing kernels.

Assessment of Winograd-based processor element for CNN processor and ASIC implementation of convolution processor element

The convolutional neural network (CNN) is the most widely used machine learning technique within the fields of image and video processing. It is primarily used to categorize images using vast datasets. This require a lot of calculations. The effectiveness of a field-programmable gate array (FPGA) as a hardware accelerator for CNNs will give excellent performance at low power budgets. The employment of the Winograd algorithm can reduce the number of processing stages in CNN. 2-D convolution is employed for the bulk of calculations in CNNs. The tactic for computing convolution for smaller filter sizes that uses Winograd minimum filtering is the handiest. The comparison of computation complexity for multiplications will be performed using Matlab. The architecture of the Winograd-based processing element, RTL coding in Verilog HDL, and the test bench are designed to examine the performance. The Xilinx Vivado / Cadence tool will be used to implement the processing element in the convolution unit on an FPGA or ASIC.

Immunity-based Ebola optimization search algorithm for minimization of feature extraction with reduction in digital mammography using CNN models

Feature classification in digital medical images like mammography presents an optimization problem which researchers often neglect. The use of a convolutional neural network (CNN) in feature extraction and classification has been widely reported in the literature to have achieved outstanding performance and acceptance in the disease detection procedure. However, little emphasis is placed on ensuring that only discriminant features extracted by the convolutional operations are passed on to the classifier, to avoid bottlenecking the classification operation. Unfortunately, since this has been left unaddressed, a subtle performance impairment has resulted from this omission. Therefore, this study is devoted to addressing these drawbacks using a metaheuristic algorithm to optimize the number of features extracted by the CNN, so that suggestive features are applied for the classification process. To achieve this, a new variant of the Ebola-based optimization algorithm is proposed, based on the population immunity concept and the use of a chaos mapping initialization strategy. The resulting algorithm, called the immunity-based Ebola optimization search algorithm (IEOSA), is applied to the optimization problem addressed in the study. The optimized features represent the output from the IEOSA, which receives the noisy and unfiltered detected features from the convolutional process as input. An exhaustive evaluation of the IEOSA was carried out using classical and IEEE CEC benchmarked functions. A comparative analysis of the performance of IEOSA is presented, with some recent optimization algorithms. The experimental result showed that IEOSA performed well on all the tested benchmark functions. Furthermore, IEOSA was then applied to solve the feature enhancement and selection problem in CNN for better prediction of breast cancer in digital mammography. The classification accuracy returned by the IEOSA method showed that the new approach improved the classification process on detected features when using CNN models.

Signal recovery in optical wireless communication using photonic convolutional processor.

Deep neural networks (DNNs) have been applied to recover signals in optical communication systems and have shown competence of mitigating linear and nonlinear distortions. However, as the data throughput increases, the heavy computational cost of DNNs impedes them from rapid and power-efficient processing. In this paper, we propose an optical communication signal recovery technology based on a photonic convolutional processor, which is realized by dispersion delay unit and wavelength division multiplexing. Based on the photonic convolutional processor, we implement an optoelectronic convolutional neural network (OECNN) for signal post-equalization and experimentally demonstrate on 16QAM and 32QAM of an optical wireless communication system. With system parameters optimization, we verify that the OECNN can achieve accurate signal recovery where the bit error ratio (BER) is below the 7% forward error correction threshold of 3.8×10-3 at 2Gbps. With adding the OECNN-based nonlinear compensation, compared with only linear compensation, we improve the quality (Q) factor by 3.35 dB at 16QAM and 3.30 dB at 32QAM, which is comparable to that of an electronic neural network. This work proves that the photonic implementation of DNN is promising to provide a fast and power-efficient solution for optical communication signal processing.

Multiple Video Frame Interpolation via Enhanced Deformable Separable Convolution.

Generating non-existing frames from a consecutive video sequence has been an interesting and challenging problem in the video processing field. Typical kernel-based interpolation methods predict pixels with a single convolution process that convolves source frames with spatially adaptive local kernels, which circumvents the time-consuming, explicit motion estimation in the form of optical flow. However, when scene motion is larger than the pre-defined kernel size, these methods are prone to yield less plausible results. In addition, they cannot directly generate a frame at an arbitrary temporal position because the learned kernels are tied to the midpoint in time between the input frames. In this paper, we try to solve these problems and propose a novel non-flow kernel-based approach that we refer to as enhanced deformable separable convolution (EDSC) to estimate not only adaptive kernels, but also offsets, masks and biases to make the network obtain information from non-local neighborhood. During the learning process, different intermediate time step can be involved as a control variable by means of an extension of coord-conv trick, allowing the estimated components to vary with different input temporal information. This makes our method capable to produce multiple in-between frames. Furthermore, we investigate the relationships between our method and other typical kernel- and flow-based methods. Experimental results show that our method performs favorably against the state-of-the-art methods across a broad range of datasets. Code will be publicly available on URL: https://github.com/Xianhang/EDSC-pytorch.