14,842 Hits in 6.2 sec

Why Are Convolutional Nets More Sample-Efficient than Fully-Connected Nets? [article]

Zhiyuan Li, Yi Zhang, Sanjeev Arora
2021 arXiv   pre-print
However, this has not been made mathematically rigorous, and the hurdle is that the fully connected net can always simulate the convolutional net (for a fixed task).  ...  Convolutional neural networks often dominate fully-connected counterparts in generalization performance, especially on image classification tasks.  ...  Du et al. (2018) attempted to investigate the reason why convolutional nets are more sample efficient.  ... 
arXiv:2010.08515v2 fatcat:suxiuchnajc7nnhsnttus5p6g4

CNN Architectures: Alex Net, Le Net, VGG, Google Net, Res Net

2020 International journal of recent technology and engineering  
Convolutional Neural Networks(CNNs) are a floating area in Deep Learning. Now a days CNNs are used inside the more note worthy some portion of the Object Recognition tasks.  ...  Right now, inside and out assessment of CNN shape and projects are built up. A relative examine of different assortments of CNN are too portrayed on this work.  ...  Fig. 7 The end Google Net created from a couple of such beginning modules stacked more than each other with coincidental pooling layers inside the focal point of, more than one more convolutional layers  ... 
doi:10.35940/ijrte.f9532.038620 fatcat:63flwu24wvhf3fsc3tdxp3hrqy

MSDU-net: A Multi-Scale Dilated U-net for Blur Detection [article]

Fan Yang, Xiao Xiao
2020 arXiv   pre-print
Inspired by the success of the U-net architecture for image segmentation, we design a Multi-Scale Dilated convolutional neural network based on U-net, which we call MSDU-net.  ...  We show that using the MSDU-net we are able to outperform other state of the art blur detection methods on two publicly available benchmarks.  ...  connection; red arrows are contracting with pooling layers; yellow arrows are expanding with up-sampling layer; blue blocks are extractors with dilated convolution; orange blocks are common convolution  ... 
arXiv:2006.03182v1 fatcat:z6zqw3o2qzechch3eijuhuu2gi

FFU-Net: Feature Fusion U-Net for Lesion Segmentation of Diabetic Retinopathy

Yifei Xu, Zhuming Zhou, Xiao Li, Nuo Zhang, Meizi Zhang, Pingping Wei, Changming Sun
2021 BioMed Research International  
Then, we integrate multiscale feature fusion (MSFF) block into the encoders which helps the network to learn multiscale features efficiently and enrich the information carried with skip connection and  ...  Firstly, the pooling layer in the network is replaced with a convolutional layer to reduce spatial loss of the fundus image.  ...  After the pooling layers are replaced with 3 × 3 convolution layers and all the activations are set to RReLu, U-Net V1 achieve slightly better than U-Net-FL.  ... 
doi:10.1155/2021/6644071 pmid:33490274 pmcid:PMC7801055 fatcat:otngnb536bfgjffr5nkri7ryba

U-Net-Based Medical Image Segmentation

Xiao-Xia Yin, Le Sun, Yuhan Fu, Ruiliang Lu, Yanchun Zhang, Hangjun Che
2022 Journal of Healthcare Engineering  
With the increasing requirements for the performance of segmentation in medical imaging in recent years, U-Net has been cited academically more than 2500 times.  ...  This paper summarizes the medical image segmentation technologies based on the U-Net structure variants concerning their structure, innovation, efficiency, etc.; reviews and categorizes the related methodology  ...  Now that 3D U-Net is widely used, why is 2D still useful?  ... 
doi:10.1155/2022/4189781 pmid:35463660 pmcid:PMC9033381 fatcat:juxw7yh2j5f5le3kjl4tkacxju

Sketch-a-Net that Beats Humans [article]

Qian Yu, Yongxin Yang, Yi-Zhe Song, Tao Xiang, Timothy Hospedales
2015 arXiv   pre-print
Our network on the other hand not only delivers the best performance on the largest human sketch dataset to date, but also is small in size making efficient training possible using just CPUs.  ...  Our superior performance is a result of explicitly embedding the unique characteristics of sketches in our model: (i) a network architecture designed for sketch rather than natural photo statistics, (ii  ...  Then two more fully connected layers are appended. Dropout regularisation [8] is applied on the first two fully connected layers.  ... 
arXiv:1501.07873v3 fatcat:vci4wbo5jrg4zn4andvijk6rpm

PERF-Net: Pose Empowered RGB-Flow Net [article]

Yinxiao Li and Zhichao Lu and Xuehan Xiong and Jonathan Huang
2021 arXiv   pre-print
Using this insight, we then propose a new model, which we dub PERF-Net (short for Pose Empowered RGB-Flow Net), which combines this new pose stream with the standard RGB and flow based input streams via  ...  In recent years, many works in the video action recognition literature have shown that two stream models (combining spatial and temporal input streams) are necessary for achieving state of the art performance  ...  We argue that the reason is because there are quite a few action training examples that are missing more than 50% of the human body; thus pose cannot be determined in such frames.  ... 
arXiv:2009.13087v2 fatcat:fzgsotztfvdc5df4gm3wjyn4gu

Prostate Segmentation using 2D Bridged U-net [article]

Wanli Chen, Yue Zhang, Junjun He, Yu Qiao, Yifan Chen, Hongjian Shi, Xiaoying Tang
2018 arXiv   pre-print
To address the aforementioned three problems, we propose and validate a deeper network that can fit medical image datasets that are usually small in the sample size.  ...  of more learnable parameters.  ...  Besides, the segmentation boundary are more continuous and smooth than the competing method.  ... 
arXiv:1807.04459v2 fatcat:c2g2g7y2k5g2xancdpuuz3uvdy

Do Deep Convolutional Nets Really Need to be Deep and Convolutional? [article]

Gregor Urban, Krzysztof J. Geras, Samira Ebrahimi Kahou, Ozlem Aslan, Shengjie Wang, Rich Caruana, Abdelrahman Mohamed, Matthai Philipose, Matt Richardson
2017 arXiv   pre-print
Although previous research showed that shallow feed-forward nets sometimes can learn the complex functions previously learned by deep nets while using the same number of parameters as the deep models they  ...  This paper provides the first empirical demonstration that deep convolutional models really need to be both deep and convolutional, even when trained with methods such as distillation that allow small  ...  Deep convolutional nets are significantly more accurate than shallow convolutional models, given the same parameter budget.  ... 
arXiv:1603.05691v4 fatcat:nixa6yw7zje5nb2rdqrmf6tqwu

Factorizable Net: An Efficient Subgraph-based Framework for Scene Graph Generation [article]

Yikang Li, Wanli Ouyang, Bolei Zhou, Jianping Shi, Chao Zhang, Xiaogang Wang
2018 arXiv   pre-print
To improve the efficiency of scene graph generation, we propose a subgraph-based connection graph to concisely represent the scene graph during the inference.  ...  All the experiments are performed on VG-MSDN [35] as it is larger than VRD [37] to eliminate overfitting and contains more predicate categories than VG-DR-Net [6] .  ...  Thus, after the pooling, 2-D convolution layers and fully-connected layers are used to transform the subgraph feature and object features respectively.  ... 
arXiv:1806.11538v2 fatcat:rn5aihsfmnfllom3dla4t5k2i4

SDGH-Net: Ship Detection in Optical Remote Sensing Images Based on Gaussian Heatmap Regression

Zhenqing Wang, Yi Zhou, Futao Wang, Shixin Wang, Zhiyu Xu
2021 Remote Sensing  
To avoid the over-fitting problem that may arise from the fully connected layer, we propose a fully convolutional neural network, SDGH-Net, based on Gaussian heatmap regression.  ...  These detection algorithms use fully connected layer direct regression to obtain coordinate points. Although training and forward speed are fast, they lack spatial generalization ability.  ...  SDGH-Net is a fully convolutional model that avoids the loss of feature map spatial information due to the fully connected layer.  ... 
doi:10.3390/rs13030499 fatcat:qh43sdf3lvcn3mmw6wgrztzgoy

Microscopy cell nuclei segmentation with enhanced U-Net

Feixiao Long
2020 BMC Bioinformatics  
Through strictly controlled experiments, the average IOU and precision of U-Net+ predictions are confirmed to outperform other prevalent competing methods with 1.0% to 3.0% gain on the first stage test  ...  An enhanced, light-weighted U-Net (called U-Net+) with modified encoded branch is proposed to potentially work with low-resources computing.  ...  For simplicity, the convolutional blocks are not shown in the up-sampling blocks. b One skip-connection of U-Net++.  ... 
doi:10.1186/s12859-019-3332-1 pmid:31914944 pmcid:PMC6950983 fatcat:vuvux3ekvjh2nj5f4kq6fj5ewi

Res-Dense Net for 3D Covid Chest CT-scan classification [article]

Quoc-Huy Trinh, Minh-Van Nguyen, Thien-Phuc Nguyen Dinh
2022 arXiv   pre-print
There are many tasks to diagnose the illness through CT-scan images, include COVID-19.  ...  In our method, we experiment with two backbones are DenseNet 121 and ResNet 101. This method achieves a competitive performance on some evaluation metrics  ...  Densely Connected Convolutional Network The demonstration of the recent work has shown that Convolutional Neural Networks can be substantially deeper, more accurate, and more efficient to train if they  ... 
arXiv:2208.04613v1 fatcat:w4hc2oqeiffybnq2zhbyxmu2h4

Network Anomaly Detection With Temporal Convolutional Network and U-Net Model

Anzhelika Mezina, Radim Burget, Carlos M. Travieso-Gonzalez
2021 IEEE Access  
so successful on more complex and actual data.  ...  In this work, we applied a couple of new methods based on convolutional neural networks: U-Net based and Temporal convolutional network based for network attack classification.  ...  To adapt this architecture to the classification task, fully connected layers are appended.  ... 
doi:10.1109/access.2021.3121998 fatcat:tnmdvdvemvev3m5sm5g3i3avqi

Automatic Lumbar Spinal CT Image Segmentation with a Dual Densely Connected U-Net [article]

He Tang, Xiaobing Pei, Shilong Huang, Xin Li, Chao Liu
2020 arXiv   pre-print
segmentation labels; 2) a dual densely connected U-shaped neural network (DDU-Net) is used to segment the spinal canal, dural sac and vertebral body in an end-to-end manner; 3) DDU-Net is capable of segmenting  ...  We expect that the technique will increase both radiology workflow efficiency and the perceived value of radiology reports for referring clinicians and patients.  ...  DDU-NET ARCHITECTURE We propose a deep fully convolutional network to segment the CT images.  ... 
arXiv:1910.09198v2 fatcat:7pf32wxwo5gx3ipdi57nh2gtby
« Previous Showing results 1 — 15 out of 14,842 results