Filters








81,067 Hits in 1.8 sec

Augmented Spatial Pooling [chapter]

John Thornton, Andrew Srbic, Linda Main, Mahsa Chitsaz
2011 Lecture Notes in Computer Science  
Specifically, we evaluate the performance of a recently proposed binary spatial pooling algorithm on a well-known benchmark of greyscale natural images.  ...  In this paper we investigate a brain-inspired spatial pooling algorithm that produces such sparse distributed representations by modelling the formation of proximal dendrites associated with neocortical  ...  Comparison of augmented spatial pooling (ASP) with binary spatial pooling (BSP) on the complete set of binary scaled natural images taken from [11] .  ... 
doi:10.1007/978-3-642-25832-9_27 fatcat:l6zos4qhwvdilnmey5dpjkmdyi

Fully Convolutional Neural Network with Augmented Atrous Spatial Pyramid Pool and Fully Connected Fusion Path for High Resolution Remote Sensing Image Segmentation

Guangsheng Chen, Chao Li, Wei Wei, Weipeng Jing, Marcin Woźniak, Tomas Blažauskas, Robertas Damaševičius
2019 Applied Sciences  
The network is built with the computationally-efficient DeepLabv3 architecture, with added Augmented Atrous Spatial Pyramid Pool and FC Fusion Path layers.  ...  (a) Augmented Atrous Spatial Pyramid Pooling (A-ASPP) module with three parallel dilated convolution branches, each branch consists of four different expansion rate dilated convolution layers.  ...  (a) Augmented Atrous Spatial Pyramid Pooling (A-ASPP) module with three parallel dilated convolution branches, each branch consists of four different expansion rate dilated convolution layers.  ... 
doi:10.3390/app9091816 fatcat:6y7ewhfq3ncodpzs7koni5dtru

Hierarchically Decoupled Spatial-Temporal Contrast for Self-supervised Video Representation Learning [article]

Zehua Zhang, David Crandall
2021 arXiv   pre-print
We show by experiments that augmentations can be manipulated as regularization to guide the network to learn desired semantics in contrastive learning, and we propose a way for the model to separately  ...  capture spatial and temporal features at multiple scales.  ...  Feature maps at multiple scales are 2D global average pooled along the spatial dimension (S-Avg Pooling) in Hierarchical Spatial Contrast learning, and 3D global average pooled along the temporal and spatial  ... 
arXiv:2011.11261v2 fatcat:aztia46gm5bzzln4i5sjfm4why

Topology-preserving augmentation for CNN-based segmentation of congenital heart defects from 3D paediatric CMR [article]

Nick Byrne, James R. Clough, Isra Valverde, Giovanni Montana, Andrew P. King
2019 arXiv   pre-print
Even when a label set has a high spatial overlap with the ground truth, inaccurate delineation of these interfaces can result in topological errors.  ...  Whilst data augmentation has often played an important role, we show that conventional image resampling schemes used therein can introduce topological changes in the ground truth labelling of augmented  ...  As well as spatial overlap, we are also concerned with the topology of the inferred blood pool label map.  ... 
arXiv:1908.08870v1 fatcat:ksgfoxistffyfmuv4y5qev6uwi

S3Pool: Pooling with Stochastic Spatial Sampling

Shuangfei Zhai, Hui Wu, Abhishek Kumar, Yu Cheng, Yongxi Lu, Zhongfei Zhang, Rogerio Feris
2017 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)  
We view the pooling operation in CNNs as a twostep procedure: first, a pooling window (e.g., 2ˆ2) slides over the feature map with stride one which leaves the spatial resolution intact, and second, downsampling  ...  We study this aspect and propose a novel pooling strategy with stochastic spatial sampling (S3Pool), where the regular downsampling is replaced by a more general stochastic version.  ...  The high performance of S3Pool even without data augmentation is consistent with our understanding of the stochastic spatial sampling step as an implicit data augmentation strategy.  ... 
doi:10.1109/cvpr.2017.426 dblp:conf/cvpr/ZhaiWKCLZF17 fatcat:socf2w6n25gr5agfwhbtwkjtdi

S3Pool: Pooling with Stochastic Spatial Sampling [article]

Shuangfei Zhai, Hui Wu, Abhishek Kumar, Yu Cheng, Yongxi Lu, Zhongfei Zhang, Rogerio Feris
2016 arXiv   pre-print
We view the pooling operation in CNNs as a two-step procedure: first, a pooling window (e.g., 2× 2) slides over the feature map with stride one which leaves the spatial resolution intact, and second, downsampling  ...  We study this aspect and propose a novel pooling strategy with stochastic spatial sampling (S3Pool), where the regular downsampling is replaced by a more general stochastic version.  ...  The high performance of S3Pool even without data augmentation is consistent with our understanding of the stochastic spatial sampling step as an implicit data augmentation strategy.  ... 
arXiv:1611.05138v1 fatcat:uinmplhbl5faxb74seyyesezeu

Fractional Max-Pooling [article]

Benjamin Graham
2015 arXiv   pre-print
Convolutional networks almost always incorporate some form of spatial pooling, and very often it is alpha times alpha max-pooling with alpha=2.  ...  However, if you simply alternate convolutional layers with max-pooling layers, performance is limited due to the rapid reduction in spatial size, and the disjoint nature of the pooling regions.  ...  . • Some form of spatial pooling, such as max-pooling.  ... 
arXiv:1412.6071v4 fatcat:xobyc2drrzfyxe3c5fzno74pbq

Few-shot Action Recognition with Permutation-invariant Attention [article]

Hongguang Zhang, Li Zhang, Xiaojuan Qi, Hongdong Li, Philip H. S. Torr, Piotr Koniusz
2020 arXiv   pre-print
Importantly, to re-weight block contributions during pooling, we exploit spatial and temporal attention modules and self-supervision.  ...  Subsequently, the pooled representations are combined into simple relation descriptors which encode so-called query and support clips.  ...  We apply the same augmentation(s) on the temporal or spatial attention vectors of the original data resulting in the augmented attention vectors which we align with attention vectors of the augmented data  ... 
arXiv:2001.03905v3 fatcat:bfj2xhgavvgete5m347gja6ney

Empirical Remarks on the Translational Equivariance of Convolutional Layers

Kyung Joo Cheoi, Hyeonyeong Choi, Jaepil Ko
2020 Applied Sciences  
In this paper, we investigate how vulnerable CNNs without pooling or augmentation are to translation in object recognition.  ...  The pooling layers provide some level of invariance. In object recognition, invariance is more important than equivariance.  ...  On the Translational Invariance of Max-Pooling and Augmentation This section demonstrates the well-known property of max-pooling layers and augmentation in terms of translational invariance.  ... 
doi:10.3390/app10093161 fatcat:hmgdm7sgbnfgfj3q7fuy2b2xl4

Spectral-spatial classification of hyperspectral imagery using a dual-channel convolutional neural network

Haokui Zhang, Ying Li, Yuzhu Zhang, Qiang Shen
2017 Remote Sensing Letters  
To overcome the problem of the limited available training samples in HSIs, we propose a simple data augmentation method which is e cient and e↵ective for improving HSI classification accuracy.  ...  In this paper, a novel dual-channel convolutional neural network (DC-CNN) framework is proposed for accurate spectral-spatial classification of hyperspectral image (HSI).  ...  Firstly, the spectral and spatial features are concatenated together as following F = [pool(F 1 1D ), F 2 1D , pool(F 1 2D ), F 2 2D ] (4) where F is the concatenated spectral-spatial features, pool()  ... 
doi:10.1080/2150704x.2017.1280200 fatcat:k2wivulba5fancunj6lobzdecm

Spatial pooling in the second-order spatial structure of cortical complex cells

Ko Sakai, Shigeru Tanaka
2000 Vision Research  
The results support the cascade mechanism consisting of simple cells' local feature extraction followed by spatial pooling.  ...  Models with nonlinear spatial pooling of simple-cell-like linear subunits reproduce the second-order kernels in good agreement with physiologically estimated kernels, while models without the pooling mechanism  ...  The third stage pools the outputs of the second stage over spatial phases and/or a spatial neighborhood.  ... 
doi:10.1016/s0042-6989(99)00230-8 pmid:10683461 fatcat:a6px45zao5bbxnlwdjr6zkznym

Parallel Grid Pooling for Data Augmentation [article]

Akito Takeki, Daiki Ikami, Go Irie, Kiyoharu Aizawa
2018 arXiv   pre-print
It works as data augmentation and is complementary to commonly used data augmentation techniques.  ...  Motivated by this observation, we propose a novel layer called parallel grid pooling (PGP) which is applicable to various CNN models.  ...  Let F s k be a spatial operation (e.g., convolution or pooling) with a kernel size of k × k and stride of s × s (s ≥ 2).  ... 
arXiv:1803.11370v1 fatcat:54d2zujifzhejiw7spuhvhnjqq

Combining Pyramid Pooling and Attention Mechanism for Pelvic MR Image Semantic Segmentaion [article]

Ting-Ting Liang, Satoshi Tsutsui, Liangcai Gao, Jing-Jing Lu and Mengyan Sun
2018 arXiv   pre-print
To address the challenges, we propose a novel convolutional neural network called Attention-Pyramid network (APNet) that effectively exploits both local and global contexts, in addition to a data-augmentation  ...  Pyramid Pooling Layer Spatial Pyramid Pooling (SPP) [8] is to gather multiple levels of local contexts by pooling with multiple kernel sizes. Zhao et al.  ...  Both APNet and PSPNet have spatial pyramid pooling [8] with the level of 4 [22] .  ... 
arXiv:1806.00264v2 fatcat:3fopjk3ysbcnhbyira3r5h4xvu

Traffic Sign Classification Using Deep Inception Based Convolutional Networks [article]

Mrinal Haloi
2016 arXiv   pre-print
augmentations.  ...  Use of spatial transformer layer makes this network more robust to deformations such as translation, rotation, scaling of input images.  ...  We have used four spatial transformer layers (networks), TABLE II SPATIAL II TRANSFORMER LAYER(NETWORKS) CONFIGURATIONS Net conv/stride max-pool conv/stride max-pool fc fc ST1 128,5x5/2 yes  ... 
arXiv:1511.02992v2 fatcat:phu5clwglrh55ow3p5v36djfx4

Batch-normalized Maxout Network in Network [article]

Jia-Ren Chang, Yong-Sheng Chen
2015 arXiv   pre-print
Finally, average pooling is used in all pooling layers to regularize maxout MLP in order to facilitate information abstraction in every receptive field while tolerating the change of object position.  ...  Because average pooling preserves all features in the local patch, the proposed MIN model can enforce the suppression of irrelevant information during training.  ...  Thus, we are able to use spatial average pooling in each pooling layer to aggregate local spatial information.  ... 
arXiv:1511.02583v1 fatcat:cldzkiis7fd3xew5222qbcxnai
« Previous Showing results 1 — 15 out of 81,067 results