Convolutional neural networks at constrained time cost
Open accessThe DOI matches its Crossref record; 26 of the 30 references it cites were found in a registry or an index.
Abstract
Though recent advanced convolutional neural networks (CNNs) have been improving the image recognition accuracy, the models are getting more complex and time-consuming. For real-world applications in industrial and commercial scenarios, engineers and developers are often faced with the requirement of constrained time budget. In this paper, we investigate the accuracy of CNNs under constrained time cost. Under this constraint, the designs of the network architectures should exhibit as trade-offs among the factors like depth, numbers of filters, filter sizes, etc. With a series of controlled comparisons, we progressively modify a baseline model while preserving its time complexity. This is also helpful for understanding the importance of the factors in network designs. We present an architecture that achieves very competitive accuracy in the ImageNet dataset (11.8% top-5 error, 10-view test), yet is 20% faster than “AlexNet” [14] (16.0% top-5 error, 10-view test).
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“The performance of neural network models is closely related to the setting of hyperparameters [42].”
From Prediction of Heat and Cold Loads of Factory Mushroom Houses Based on EWT Decomposition (2023) · cites this paper as [42]
Citations
Cites
Very Deep Convolutional Networks for Large-Scale Image Recognition
Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148 · cited 9×
“No other data augmentation, such as scale jittering [10, 22, 9], is used.”
BackgroundMethodInfluentialImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155 · cited 14×
“The actual running time of our model is 20% faster than AlexNet.”
BackgroundMethodResultInfluentialImageNet: A large-scale hierarchical image database
Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., & Fei-Fei, L. (2009) · 2009 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2009.5206848 · Cited by 65,333 · cited 7×
“Through the investigations, we obtain a model that achieves 11.8% top-5 error (10-view test) on ImageNet [3] and only takes 3 to 4 days training on a single GPU.”
BackgroundMethodInfluentialGoing deeper with convolutions
Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., Erhan, D., Vanhoucke, V., & Rabinovich, A. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7298594 · Cited by 48,562
ImageNet Large Scale Visual Recognition Challenge
Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., et al. (2015) · International Journal of Computer Vision · doi:10.1007/s11263-015-0816-y · Cited by 39,380
Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation
Girshick, R., Donahue, J., Darrell, T., & Malik, J. (2014) · 2014 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2014.81 · Cited by 32,684
(2010) · Cited by 14,134
(2010) · International Conference on Machine Learning · Cited by 13,995
Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2015) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2015.2389824 · Cited by 11,971
Self-citationBackpropagation Applied to Handwritten Zip Code Recognition
LeCun, Y., Boser, B., Denker, J. S., Henderson, D., Howard, R. E., Hubbard, W., & Jackel, L. D. (1989) · Neural Computation · doi:10.1162/neco.1989.1.4.541 · Cited by 11,891 · cited 1×
“Convolutional neural networks (CNNs) [15, 14] have recently brought in revolutions to the computer vision area.”
Background
Research works printed but not linked: 1
- lin, 2013, Network in network, arXiv 1312 4400
Cited by
Items 901–1,000 of 1,001
Yuan, H., Song, F., Chu, X., Li, W., Wang, X., Han, H., & Gong, Y. (2021) · IET Communications · doi:10.1049/cmu2.12257 · Cited by 10
MTMI-DCNN: A PSR-Based Method for Time Series Sensor Data Classification
Liu, R., Wang, F., Tian, F., Qian, J., Chen, X., Cui, S., & Lin, Y. (2022) · IEEE Sensors Journal · doi:10.1109/jsen.2022.3149948 · Cited by 10
DeepAntiJam: Stackelberg Game-Oriented Secure Transmission via Deep Reinforcement Learning
Lu, J., He, D., & Wang, Z. (2022) · IEEE Communications Letters · doi:10.1109/lcomm.2022.3182001 · Cited by 10
“2The time complexity of our considered solver NN is O(2n1 + n1n2 + n2), whereby ni denotes the nodes number of i-th layer [15], while the complexity of bisection method is O log2(NPA ) , where NPA is the size of available transmit power set.”
BackgroundDomain adversarial tangent subspace alignment for explainable domain adaptation
Raab, C., Röder, M., & Schleif, F.-M. (2022) · Neurocomputing · doi:10.1016/j.neucom.2022.07.074 · Cited by 12
Copy-Move Forgery Detection in Digital Forensic Images Using CNN
Kaya, M., Sani, K. J., & Karakuş, S. (2022) · International Conference on Computer Science and Engineering (UBMK) · doi:10.1109/ubmk55850.2022.9919560 · Cited by 10
Conditional Independence Induced Unsupervised Domain Adaptation
Xu, X., Xu, G.-X., Ren, C.-X., Dai, D., & Yan, H. (2023) · Pattern Recognition · doi:10.1016/j.patcog.2023.109787 · Cited by 10
Kang, Z., Hei, B., Zhou, Z., & Li, S. (2018) · National Remote Sensing Bulletin · doi:10.11834/jrs.20187075 · Cited by 10
Can You Tell a Face from a HEVC Bitstream?
Alvar, S. R., Choi, H., & Bajić, I. V. (2018) · IEEE International Conference on Multimedia Information Processing and Retrieval (MIPR) · doi:10.1109/mipr.2018.00060 · Cited by 16
Navpreet, Roul, R. K., & Rani, R. (2024) · Multimedia Tools and Applications · doi:10.1007/s11042-024-19291-9 · Cited by 9
“According to He et al. [48], the fully connected layer takes 10% of computation time in CNN, leading to a higher duration.”
Ke, H., Chen, Z., Fan, X., Yang, C., & Wang, H. (2024) · ISA Transactions · doi:10.1016/j.isatra.2024.08.019 · Cited by 11
Redundant Feature-Processing Module Based on Dual-Branch for Underwater Image Enhancement
Zhang, Y., Li, Z., & Namiki, A. (2024) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2024.3480228 · Cited by 9
“1) Performance Saturation and Degradation: From He and Sun’s [21] proof, it is clear that deeper is not always better, and the performance of the network not only gets saturated at some point but may worsen if going deeper.”
BackgroundChen, D., Zhang, Z., Zhou, F., & Wang, C. (2024) · Entropy · doi:10.3390/e26121007 · Cited by 9
Osmani, N., Esmaeeli, E., & Rezayi, S. (2025) · The Journal of Supercomputing · doi:10.1007/s11227-024-06663-z · Cited by 9
“The neural network, comprising 60 million parameters and 650,000 neurons, features five convolutional layers, some accompanied by max-pooling layers, and three fully connected layers culminating in a final softmax layer [30, 31].”
Learning Sparse Filters in Deep Convolutional Neural Networks with a $$l_1/l_2$$ Pseudo-Norm
Berthelier, A., Yan, Y., Château, T., Blanc, C., Duffner, S., & García, C. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-68763-2_50 · Cited by 12
Machine Learning Approach for Accelerating Simulation-based Fault Injection
Lu, L., Chen, J., Breitenreiter, A., Schrape, O., Ulbricht, M., & Krstic, M. D. (2021) · IEEE Nordic Circuits and Systems Conference (NorCAS) · doi:10.1109/norcas53631.2021.9599646 · Cited by 9
Benchmark Analysis of Semantic Segmentation Algorithms for Safe Planetary Landing Site Selection
Claudet, T., Tomita, K., & Ho, K. (2022) · IEEE Access · doi:10.1109/access.2022.3167763 · Cited by 10
“They are based on the 2D cross-correlation operation and lead, following [38], for squared input and kernel, to : with C t the time complexity, l the index of the layer, n l − 1 the number of input channels of layer l, n l the width, or number of filters, s l the spatial size of the filter, and m l ,…”
BackgroundGuo, W., Liu, S., Gong, Z., Zhang, G., & Jiang, X. (2022) · International Journal of Imaging Systems and Technology · doi:10.1002/ima.22790 · Cited by 9
Zhou, Y. Z., Xia, L., Zhao, J., Yao, R., & Liu, B. (2023) · Multimedia Tools and Applications · doi:10.1007/s11042-023-15608-2 · Cited by 15
“Conventional methods include using fewer channels and adding more layers appropriately [71].”
SweepNet: A Lightweight CNN Architecture for the Classification of Adaptive Genomic Regions
Zhao, H., Pavlidis, P., & Alachiotis, N. (2023) · Platform for Advanced Scientific Computing Conference (PASC) · doi:10.1145/3592979.3593411 · Cited by 9
“Unlike summary statistics though, which rely on the underlying population genetics theory, a CNN abstracts the implicit information within an image through various aspects like the topological structure and high dimensional features [26].”
BackgroundLi, M., Wang, Z., Zhang, Y., & Gao, P. (2023) · IEEE Sensors Journal · doi:10.1109/jsen.2023.3336307 · Cited by 10
“The research finding is that during training, as the model depth increases, the model accuracy tends to saturate and then decreases [23], [24].”
BackgroundCui, Y., & Feng, Y. (2025) · International Journal of Electrochemical Science · doi:10.1016/j.ijoes.2025.101008 · Cited by 9
Hu, W., Jin, J., Liu, T., & Zhang, C. (2019) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2019.2939157 · Cited by 11
“Referring to [46], we calculate the number of required operations (convolution, nonlinear, and full connection).”
BackgroundSankaran, D. G., Stemm-Wolf, A. J., McCurdy, B. L., Hariharan, B., & Pearson, C. G. (2020) · Journal of Cell Science · doi:10.1242/jcs.243543 · Cited by 9
Gated residual neural networks with self-normalization for translation initiation site recognition
Guo, Y., Zhou, D., Cao, J., Nie, R., Ruan, X., & Liu, Y. (2021) · Knowledge-Based Systems · doi:10.1016/j.knosys.2021.107783 · Cited by 9
Channel Estimation via Model and Learning for Monostatic Multiantenna Backscatter Communication
Yerzhanova, M., & Kim, Y. H. (2021) · IEEE Access · doi:10.1109/access.2021.3134961 · Cited by 9
“A time complexity of a CNN-based network was computed as [35] where L is the number of convolution layers, n w,l and n h,l are the width and height of the l th layer, s 2 l is the kernel size, and f l − 1 and f l are the number of input and output feature maps (channels), respectively.”
BackgroundShip Detection in Complex Environment Using SAR Time Series
Kahar, S., Hu, F., & Xu, F. (2022) · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · doi:10.1109/jstars.2022.3170361 · Cited by 9
Sustainable Oil Palm Resource Assessment Based on an Enhanced Deep Learning Method
Liu, X., Ghazali, K. H., & Shah, A. A. (2022) · Energies · doi:10.3390/en15124479 · Cited by 10
Repeatable adaptive keypoint detection via self-supervised learning
Yan, P., Tan, Y., & Tai, Y. (2022) · Science China Information Sciences · doi:10.1007/s11432-021-3364-5 · Cited by 9
End-to-End Moving Target Indication for Airborne Radar Using Deep Learning
Gu, Y., Wu, J., Fang, Y., Zhang, L., & Zhang, Q. (2022) · Remote Sensing · doi:10.3390/rs14215354 · Cited by 9
“For the mentioned D2CNN, the component complexity formula is as follows [34]:”
MethodBilliards Sports Analytics: Datasets and Tasks
Zhang, Q., Wang, Z., Long, C., & Yiu, S. (2024) · ACM Transactions on Knowledge Discovery from Data · doi:10.1145/3686804 · Cited by 10
A novel method for lung masses detection and location based on deep learning
Li, Z., & Li, L. (2017) · IEEE International Conference on Bioinformatics and Biomedicine (BIBM) · doi:10.1109/bibm.2017.8217787 · Cited by 9
“Unexpectedly, such degradation is not caused by overfitting, and adding more layers to a suitably deep model leads to higher training error, as reported in [28], [29], the author He etc.”
BackgroundAn Efficient and Resource-Aware Hashtag Recommendation Using Deep Neural Networks
Kao, D., Lai, K.-T., & Chen⋆, M.-S. (2019) · Lecture notes in computer science · doi:10.1007/978-3-030-16145-3_12 · Cited by 9
ZipNet: ZFNet-level Accuracy with 48× Fewer Parameters
Antioquia, A. M. C., Tan, D. S., Azcarraga, A. P., Cheng, W.-H., & Hua, K. (2018) · IEEE International Conference on Visual Communications and Image Processing (VCIP) · doi:10.1109/vcip.2018.8698672 · Cited by 9
“In the experiments of [8], delaying the downsampling operation reduced the top-5 errors of four CNN architectures from 14.”
MethodA Deep Learning Approach for Real-Time 3D Human Action Recognition from Skeletal Data
Pham, H., Salmane, H., Khoudour, L., Crouzil, A., Zegers, P., & Velastín, S. A. (2019) · Lecture notes in computer science · doi:10.1007/978-3-030-27202-9_2 · Cited by 9
Household Electricity Power Consumption Prediction Using CNN-GRU Techniques
Suresh, M., Anbarasi, M. S., Divyabharathi, J., Harshavardeni, D., & Meena, S. (2021) · International Conference on System, Computation, Automation and Networking (ICSCAN) · doi:10.1109/icscan53069.2021.9526485 · Cited by 9
“M. Alhussein et al, in [9], used a hybrid deep learning technique namely, the combination of CNN and LSTM, and obtained 40.38% average Mean Absolute Percentage Error (MAPE) for an individual household energy consumption.”
BackgroundMethodInfluentialHarshvardhan, G. M., Sahu, A., Gourisaria, M. K., Singh, P. K., Hong, W., Singh, V., & Balabantaray, B. K. (2022) · IEEE Access · doi:10.1109/access.2022.3159700 · Cited by 10
Li, B., & Hu, X. (2018) · Journal of Intelligent & Fuzzy Systems · doi:10.3233/jifs-17592 · Cited by 9
Effect of Successive Convolution Layers to Detect Gender
(2018) · Iraqi Journal of Science · doi:10.24996/ijs.2018.59.3c.17 · Cited by 11
“Two issues making the practical use of the Convolutional Neural Network complicated in some of the situations [31] 1) It is related to the big size of the training data that is usually needed for training them.”
BackgroundAnalysis on Gradient Propagation in Batch Normalized Residual NetworksPreprint
Panigrahi, A., Chen, Y., & Kuo, C. ‐. J. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1812.00342 · Cited by 15
“Such degradation is not caused by over-fitting since adding more layers to a suitably deep model often results in higher training errors (Srivastava et al., 2015; He and Sun, 2015).”
BackgroundTeaching Economics to the MachinesPreprint
Chen, H., Cheng, Y., Liu, Y., & Tang, K. (2023) · SSRN Electronic Journal · doi:10.2139/ssrn.4642167 · Cited by 9
Prediction of Heat and Cold Loads of Factory Mushroom Houses Based on EWT Decomposition
Zuo, H., Zheng, W., Wang, M., & Zhang, X. (2023) · Sustainability · doi:10.3390/su152115270 · Cited by 8
Global Convolutional Self-Action Module for Fast Brain Tumor Image Segmentation
Yang, W.-A., Lautan, D., Weng, T.-W., Lin, W.-C., Kao, Y., & Chen, C.-C. (2024) · IEEE Transactions on Emerging Topics in Computational Intelligence · doi:10.1109/tetci.2024.3375075 · Cited by 9
Risk-Aware Antenna Selection for Multiuser Massive MIMO Under Incomplete CSI
He, K., Vu, T. X., Hoang, D. T., Nguyen, D. N., Chatzinotas, S., & Ottersten, B. (2024) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2024.3377733 · Cited by 12
“Then, the computational complexity of CNN is given by O ( [39].”
MethodSelf-citationMukadam, S. B., & Patil, H. Y. (2024) · IEEE Access · doi:10.1109/access.2024.3457315 · Cited by 12
“Moreover, it is not cost-effective to analyze high-quality histology pictures directly using GPU memory [13].”
BackgroundAttention Map Is All We Need for Lightweight Fingerprint Liveness Detection
Yook, H. J., Hong, P. M., Kang, S. H., Jhun, G. S., Seo, J. E., & Lee, Y. K. (2024) · IEEE Access · doi:10.1109/access.2024.3458908 · Cited by 8
“However, increasing the model size (e.g., number of layers, channels, and parameters) to enhance accuracy may result in longer delay times [21], [22], [23], [24], [25].”
BackgroundNovel methodology for apple leaf disease classification with PCNN-IELM
Navpreet, & Roul, R. K. (2024) · Neural Computing and Applications · doi:10.1007/s00521-024-10816-9 · Cited by 9
“According to He et al., [24], the fully connected layers might only account for around 10% of the actual computational operations; their impact on the training time stems from the high parameter count and the intensive nature of operations required for learning and updating these parameters.”
Fan, H., Zhang, T., Cao, X., & Zhang, X. (2025) · Measurement Science and Technology · doi:10.1088/1361-6501/add038 · Cited by 9
“Although more complex features can be extracted through the deeper network, there is a degradation problem when the network layers are stacked to a certain extent [33].”
Chen, J., & Er, M. J. (2025) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2025.111067 · Cited by 10
RIS-Assisted Beamfocusing in Near-Field IoT Communication Systems: A Transformer-Based Approach
Zhou, Q., Zhao, J., Cai, K., & Zhu, Y. (2025) · IEEE Internet of Things Journal · doi:10.1109/jiot.2025.3573717 · Cited by 8
“In this section, we present a detailed analysis of the computational complexity associated with the proposed Transformer-based near-field beam training algorithm, focusing on both time complexity and memory consumption, which are quantified using floating-point operations (FLOPs) [37].”
PTEENet: Post-Trained Early-Exit Neural Networks Augmentation for Inference Cost Optimization
Lahiany, A., & Aperstein, Y. (2022) · IEEE Access · doi:10.1109/access.2022.3187002 · Cited by 15
“However, processing inputs using deeper and more complex neural network requires a significant amount of computing power and may increase the overall processing latency [16][17].”
BackgroundTeng, W. J., Teh, J. S., & Jamil, N. (2023) · Journal of Information Security and Applications · doi:10.1016/j.jisa.2023.103531 · Cited by 8
Li, W., Liu, L., Chen, Y., Xiao, Z., Le, H., & Zhang, R. (2023) · Space Weather · doi:10.1029/2022sw003376 · Cited by 9
“He & Sun, 2015; Simonyan & Zisserman, 2014).”
BackgroundInfluentialEnsemble of binary tree structured deep convolutional network for image classification
Lee, J.-E., Min-Joo, K., & Kang, J. (2017) · Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) · doi:10.1109/apsipa.2017.8282260 · Cited by 8
“and nonlinear operations are enhanced [7]–[10].”
Processor Free Time Forecasting Based on Convolutional Neural Network
Zhang, Y., Zhang, X., Cao, J., & Hui, S. (2018) · Chinese Control Conference (CCC) · doi:10.23919/chicc.2018.8483132 · Cited by 8
“The shape and number of convolution kernel directly determine the overall performance of the network [12], [13].”
BackgroundUpgrading security and protection in ear biometrics
Sinha, H., & Ajmera, P. K. (2019) · IET Biometrics · doi:10.1049/iet-bmt.2018.5081 · Cited by 8
“Recently, a general formula for the total time complexity of all convolutional layers in a CNN was presented by He and Sun [68].”
BackgroundLi, F., Yuan, D., Zhang, M., Liang, C., Zhou, X., & Zhang, H. (2019) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn.2019.8851904 · Cited by 9
“Complexity Analysis Here we quantitatively analyzed the computational and space complexity [18,19] of four architectures to find the bottleneck of the architecture.”
MethodLearning Deep Graph Representations via Convolutional Neural Networks
Ye, W., Askarisichani, O., Jones, A., & Singh, A. K. (2020) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2020.3014089 · Cited by 8
“The time complexity of two-dimensional CNNs is O (∑c l=1 nic · s(2)l · nf ·m(2)oc ) [36], where l is the index of a convolutional layer, c is the number of convolutional layers, nic is the number of input channels, sl is the length of the”
BackgroundFLaPS: Federated Learning and Privately Scaling
Paul, S., Sengupta, P., & Mishra, S. (2020) · IEEE International Conference on Mobile Adhoc and Sensor Systems · doi:10.1109/mass50613.2020.00011 · Cited by 6
“Time 3 Average training time per cluster and generating differentially private report of the weights + Training per cluster [15] O(dc)”
BackgroundLotus: A New Topology for Large-scale Distributed Machine Learning
Lu, Y., Gu, H., Yu, X., & Chakrabarty, K. (2020) · ACM Journal on Emerging Technologies in Computing Systems · doi:10.1145/3415749 · Cited by 12
“We used the formula mentioned in Reference [20] to estimate the forward propagation time, as shown below: where d is the number of convolutional layers in the neural network, l is the l th convolutional layer.”
MethodImproved Convolutional Neural Network Fault Diagnosis Method Based on Dropout
Tingting, C., Jianlin, X., & Cai, H. (2020) · International Forum on Electrical Engineering and Automation (IFEEA) · doi:10.1109/ifeea51475.2020.00160 · Cited by 8
A coarse to fine network for fast and accurate object detection in high‐resolution images
Guo, Y., Zou, Q., & Jin, L. (2021) · IET Computer Vision · doi:10.1049/cvi2.12042 · Cited by 8
“To reduce processing time in HR images, previous research [13–16] tried to simplify the network architecture or restrict the input image size to reduce GPU memory consumption and speed detection.”
MethodDeep semantic hashing with dual attention for cross-modal retrieval
Wu, J., Weng, W., Fu, J., Liu, L., & Hu, B. (2021) · Neural Computing and Applications · doi:10.1007/s00521-021-06696-y · Cited by 10
“The time complexity of the network model with multiple convolutional layers is given by [37]:”
BackgroundA Convolutional Neural Network Combined with Attributed Scattering Centers for SAR ATR
Zhou, Y., Li, Y., Xie, W., & Li, L. (2021) · Remote Sensing · doi:10.3390/rs13245121 · Cited by 8
Deep-learning approaches for pixel-level pansharpening
Yang, Y., Su, Z. Y., HUANG, S., Wei-guo, W., Tu, W., & Lu, H. (2022) · National Remote Sensing Bulletin · doi:10.11834/jrs.20211325 · Cited by 8
Robust Deep Reinforcement Learning for Interference Avoidance in Wideband Spectrum
Aref, M. A., & Jayaweera, S. K. (2019) · IEEE Cognitive Communications for Aerospace Applications Workshop (CCAAW) · doi:10.1109/ccaaw.2019.8904887 · Cited by 8
“According to [11] the pooling and the fully connected layers often take only 5–10% of the computational time.”
BackgroundMethodCategorizing License Plates Using Convolutional Neural Network with Residual Learning
Jose, J. A. C., Maningo, J. M. Z., Rogelio, J. P., Bandala, A. A., Vicerra, R. R. P., Sybingco, E., Ching, P. M. L., & Dadios, E. P. (2019) · Asia-Pacific Conference on Intelligent Robot Systems (ACIRS) · doi:10.1109/acirs.2019.8935997 · Cited by 8
“Different computer vision algorithms and image processing techniques can be used to detect and classified license plate and recognize its contents [15-22] for instance the deep learning and Convolutional Neural Networks (CNNs).”
BackgroundBalaji, K., Lavanya, K., V, & Amalanathan, G. M. (2020) · Chemometrics and Intelligent Laboratory Systems · doi:10.1016/j.chemolab.2020.104070 · Cited by 8
Deep Learning-Assisted Detection of PUE and Jamming Attacks in Cognitive Radio Systems
Aygül, M. A., Furqan, H. M., Nazzal, M., & Arslan, H. (2020) · IEEE Vehicular Technology Conference · doi:10.1109/vtc2020-fall49728.2020.9348579 · Cited by 10
“The theoretical time complexity for 1 D-CNN [19] can be given as where the number of convolutional layers is represented by b , conventional layer index is represented by l , q l represents the total number of filters used in l th layer, the length of filter is represented by ρ i , and the size of…”
BackgroundMethodImprovement on Deep Features through Various Enhancement Techniques for Vehicles Classification
Kiran, V. K., Dash, S., & Parida, P. (2021) · Sensing and Imaging · doi:10.1007/s11220-021-00363-1 · Cited by 8
“have reported that overfitting does not induce this degradation, and the insertion of additional layers into a properly deep model contributes to an increased training error [22, 23].”
BackgroundEjection Fraction estimation using deep semantic segmentation neural network
Alam, M. G. R., Khan, A. M., Shejuty, M. F., Zubayear, S. I., Shariar, M. N., Altaf, M. M., Hassan, M. M., AlQahtani, S. A., & Alsanad, A. (2022) · The Journal of Supercomputing · doi:10.1007/s11227-022-04642-w · Cited by 8
“Moreover, adding more NN layers to a very deep traditional NN model leads to higher training error, as reported in [21] and thoroughly proved by experiments.”
EV-VGCNN: A Voxel Graph CNN for Event-based Object ClassificationPreprint
Deng, Y., Chen, H., Chen, H., & Li, Y. (2021) · arXiv (Cornell University) · Cited by 10
Integration between constrained optimization and deep networks: a survey
Bizzarri, A., Fraccaroli, M., Lamma, E., & Riguzzi, F. (2024) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2024.1414707 · Cited by 8
Han, H., Xu, Y., Li, W., Wang, X., Xu, Y., Zhang, X., & Gao, Y. (2024) · IEEE Internet of Things Journal · doi:10.1109/jiot.2024.3459935 · Cited by 7
“The complexity is measured in floating-point operations (FLOPs) [45].”
Multi-Modal Image and Radio Frequency Fusion for Optimizing Vehicle Positioning
Huan, O., Luo, T., & Chen, M. (2024) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2024.3469252 · Cited by 11
“Hence, according to [34] and [36], the computational complexity of meta-learning is”
BackgroundA Novel Fast Far-Field Phased Array Calibration Method Utilizing Deep Residual Neural Networks
Chen, H., Xie, X., Ma, Z., Xu, H., Lan, B., Li, N., Qi, X., Men, C., Song, C., & Xu, Z. (2025) · IEEE Transactions on Antennas and Propagation · doi:10.1109/tap.2025.3547915 · Cited by 7
“It is known that the accuracy of networks tends to saturate and then degrades rapidly when adding more layers to a sufficiently deep model [34].”
Comprehensive Review of Deep Learning Approaches for Single-Image Super-Resolution
Liu, Z. Y., Jiang, S., Feng, S., Song, Q., & Zhang, J. (2025) · Sensors · doi:10.3390/s25185768 · Cited by 6
“However, directly stacking too many layers may lead to problems such as “gradient vanishing” and “gradient explosion” [24].”
Lu, Y., Bu, Y., Li, B., Dai, Y., Wang, S., Shi, L., & Ma, B. (2025) · Journal of Water Process Engineering · doi:10.1016/j.jwpe.2025.109050 · Cited by 7
Hou, Y., Bai, Y., Xu, T., Yan, H., Yan, H., Du, D., & Guo, Y. (2019) · Journal of Electronic Imaging · doi:10.1117/1.jei.28.5.053002 · Cited by 7
Wang, H., Zhou, Z., Yan, R., & Zhang, L. (2023) · IEEE International Instrumentation and Measurement Technology Conference · doi:10.1109/i2mtc53148.2023.10175992 · Cited by 8
“4) ResNet unit: To avoid performance degradation caused by the network deepening, the residual network (ResNet) [19] block is used in the high layer of MCAN.”
MethodDeep Learning-Assisted Channel Estimation in Frequency Selective UWA Communication Systems
Nguyễn, V. Đ., Phan, D. K., & Chien, T. V. (2023) · IEEE Access · doi:10.1109/access.2023.3311319 · Cited by 7
“Due to [43], the complexity of the convolutional layers can be calculated as O( ∑ l∈L̃ xlylk 2 l zlzl−1).”
BackgroundZeng, K., & Paik, I. (2023) · Expert Systems with Applications · doi:10.1016/j.eswa.2023.122196 · Cited by 7
Zhao, Y., Ge, P., Bi, W., & Zheng, J. (2023) · Journal of Manufacturing Processes · doi:10.1016/j.jmapro.2023.10.023 · Cited by 8
Enhanced Detection of Autism Spectrum Disorder Using ResNet: A Deep Learning Approach
Priya, S. B., Priyanga, S., & Kamisetty, S. S. S. (2024) · International Conference on Data Science and Network Security (ICDSNS) · doi:10.1109/icdsns62112.2024.10690847 · Cited by 9
“Residual Networks have achieved state-of-the-art performance on many computer vision tasks such as image classification and object detection [15].”
BackgroundFully Asynchronous Neuromorphic Perception for Mobile Robot Dodging With Loihi Chips
Jiang, J., Kong, D., Hu, C., & Fang, Z. (2025) · IEEE Transactions on Automation Science and Engineering · doi:10.1109/tase.2025.3547338 · Cited by 7
“3) Time Complexity Analysis and Inference Time: In this section, we first analyze the time complexity of di ff erent schemes according to [44].”
Shallow SqueezeNext: An Efficient & Shallow DNN
Duggal, J. K., & El‐Sharkawy, M. (2019) · IEEE International Conference on Vehicular Electronics and Safety (ICVES) · doi:10.1109/icves.2019.8906416 · Cited by 8
Using convolutional neural network (CNN) approach for ship detection in Sentinel-1 SAR imagery
Tanveer, H., Balz, T., & Mohamdi, B. (2019) · IEEE Asia-Pacific Conference on Synthetic Aperture Radar · doi:10.1109/apsar46974.2019.9048499 · Cited by 7
Going Deeper With Brain Morphometry Using Neural Networks
Cruz, R. S., Lebrat, L., Bourgeat, P., Doré, V., Dowling, J., Fripp, J., Fookes, C., & Salvado, O. (2021) · IEEE International Symposium on Biomedical Imaging (ISBI) · doi:10.1109/isbi48211.2021.9434039 · Cited by 7
“First and foremost, increasing the size of VGG results in longer training times and a more complicated optimization process [18]; Secondly, beyond a certain depth a VGG-type architecture yields less accurate results (see [19] and discussion therein).”
BackgroundExpectile regression via deep residual networks
Yin, Y., & Zou, H. (2020) · Stat · doi:10.1002/sta4.315 · Cited by 7
“The training of deeper NN can be very challenging (He & Sun, 2015; Srivastava, Greff, & Schmidhuber, 2015).”
BackgroundModel for Detection of Masquerade Attacks Based on Variable-Length Sequences
Barseghyan, G., Yuan, Y., & Anakpa, M. (2020) · IEEE Access · doi:10.1109/access.2020.3039166 · Cited by 7
“Each back-ward pass needs B = 2 × F ops, as argued by He and Sun [37].”
BackgroundA RD-T Network for Hand Gesture Recognition Based on Millimeter-Wave Sensor
Huang, R., Li, Z., Wang, S., Wang, R., Li, J., & Xu, Z. (2020) · IEEE International Conference on Signal and Image Processing (ICSIP) · doi:10.1109/icsip49896.2020.9339325 · Cited by 7
“In order to better characterize the time complexity of network, we introduce the time complexity calculation formula of convolutional neural networks [6]:”
MethodCaesarendra, W., Rahmaniar, W., Mathew, J., & Thien, A. (2022) · Lecture notes in electrical engineering · doi:10.1007/978-981-19-1804-9_41 · Cited by 7
Chen, J., Wang, Y., Zhang, L., Liu, M., & Plaza, A. (2022) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2022.3187187 · Cited by 7
“In [39], it discussed how to find a trade-off between network depth, filter size and number at constrained time cost.”
BackgroundYang, K., Yu, Z., Gu, F., Zhang, Y., Wang, S., Peng, B., & Hu, Z. (2022) · Foods · doi:10.3390/foods11203268 · Cited by 7
“Although a CNN can be used in many image recognition tasks, its large model files and high computational load lead to slow object detection [48].”
MethodIntelligent temporal detection network for boundary-sensitive flight regime recognition
Hu, C., Wu, J., Sun, C., Chen, X., & Yan, R. (2023) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2023.106949 · Cited by 7
From NeurODEs to AutoencODEs: A mean-field control framework for width-varying neural networks
Cipriani, C., Fornasier, M., & Scagliotti, A. (2024) · European Journal of Applied Mathematics · doi:10.1017/s0956792524000032 · Cited by 9
“In the practice of Deep Learning, this novel kind of layer has turned out to be highly beneficial, since it is effective in avoiding the “vanishing of the gradients” during the training [5], or the saturation of the network’s accuracy [28].”
BackgroundEncryption scheme classification: a deep learning approach
Pan, J. (2017) · International Journal of Electronic Security and Digital Forensics · doi:10.1504/ijesdf.2017.087397 · Cited by 7
Deep Convolutional Network Based on Pyramid Architecture
Lv, E., Wang, X., Cheng, Y., & Yu, Q. (2018) · IEEE Access · doi:10.1109/access.2018.2860785 · Cited by 7
“However, with the network depth increasing, accuracy gets saturated and then degrades rapidly [16]–[18].”
BackgroundNie, M., Zhang, J., & Zhang, X. (2019) · Lecture notes in computer science · doi:10.1007/978-3-030-22808-8_39 · Cited by 7
SAN : Scale-Space Attention Networks
Garg, Y., Candan, K. S., & Sapino, M. L. (2020) · IEEE International Conference on Data Engineering (ICDE) · doi:10.1109/icde48307.2020.00079 · Cited by 7
“Despite these advances, traditional DNNs are still faced with the problem of performance degradation with the increase in the depth [18]: these networks tend to saturate after a certain depth and networks suffer from limited generalization of input data into a fixed-length encoding [22].”
BackgroundGuo, H. L., Huang, S., Huang, C., Shi, F., Zhang, M., & Pan, Z. (2020) · Security and Communication Networks · doi:10.1155/2020/8881760 · Cited by 8
“In the same training round, a network with a deeper network (degraded network) has a higher error rate than a network with fewer layers [41].”
Background
Metadata
- Authors
- He, K., & Sun, J.
- Affiliations
- Microsoft (United States) (US)
- ORCID
- Venue
- 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)
- Publisher
- IEEE
- Published
- 2015-06
- Pages
- 5353-5360
- ISSN
- 1063-6919 (ISSN-L)
- OpenAlex
- W1932847118
- Cited by
- 1,558
- References
- 30
- Citations by year
- 2026: 32 · 2025: 91 · 2024: 143 · 2023: 174 · 2022: 216 · 2021: 252 · 2020: 227 · 2019: 209 · 2018: 110 · 2017: 61 · 2016: 32 · 2015: 11
- FWCI
- 29.25
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Image and Video Retrieval Techniques
- Keywords
- convolutional neural networks, time complexity, inference speed, network architecture design, speed-accuracy trade-off, ImageNet classification, AlexNet, model compression
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.