Filters








20,828 Hits in 3.1 sec

Time-aware Large Kernel Convolutions [article]

Vasileios Lioutas, Yuhong Guo
2020 arXiv   pre-print
In this paper, we introduce Time-aware Large Kernel (TaLK) Convolutions, a novel adaptive convolution operation that learns to predict the size of a summation kernel instead of using a fixed-sized kernel  ...  Alternatively, they utilize depthwise convolutions with softmax normalized kernels of size $k$ acting as a limited-window self-attention, resulting in time complexity of $O(k{\cdot}n)$.  ...  Figure 1 illustrates the Time-aware Large Kernel Convolution operation for a specific time-step during encoding.  ... 
arXiv:2002.03184v2 fatcat:2zrr75sywrdq7et737syou64zm

Factorizing and Reconstituting Large-Kernel MBConv for Lightweight Face Recognition

Yaqi Lyu, Jing Jiang, Kun Zhang, Yilun Hua, Miao Cheng
2019 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW)  
This paper investigates that factorization and reconstitution can promote the efficiency of large-kernel MBConv and thus proposes FR-MBConv (Factorizing and Reconstituting large-kernel MBConv).  ...  In addition, from the perspective of feature generation mechanism, FR-MBConv can be equivalent to more regular convolutions.  ...  In MobileNetV1 block, the time cost of the 1x1 convolution is 4.3x times than the 3x3 depthwise convolutions when testing on an Apple iPhone X [5] .  ... 
doi:10.1109/iccvw.2019.00329 dblp:conf/iccvw/LyuJZHC19 fatcat:zodpnifw5vbxtd6hscbtxigf3a

Summarize and Search: Learning Consensus-aware Dynamic Convolution for Co-Saliency Detection [article]

Ni Zhang and Junwei Han and Nian Liu and Ling Shao
2021 arXiv   pre-print
In this paper, we propose a novel consensus-aware dynamic convolution model to explicitly and effectively perform the "summarize and search" process.  ...  Next, we generate dynamic kernels from consensus features to encode the summarized consensus knowledge.  ...  Object Searching via Dynamic Convolution After obtaining consensus-aware dynamic kernels, we adopt dynamic convolution on the original feature maps {X n } N n=1 to perform explicit object searching.  ... 
arXiv:2110.00338v1 fatcat:zbvqmm5ovff65dusjf4m7tv45a

CSPN++: Learning Context and Resource Aware Convolutional Spatial Propagation Networks for Depth Completion [article]

Xinjing Cheng, Peng Wang, Chenye Guan, Ruigang Yang
2019 arXiv   pre-print
In this paper, we propose CSPN++, which further improves its effectiveness and efficiency by learning adaptive convolutional kernel sizes and the number of iterations for the propagation, thus the context  ...  In our experiments, we find weighted assembling can lead to significant accuracy improvements, which we referred to as "context-aware CSPN", while weighted selection, "resource-aware CSPN" can reduce the  ...  Resource Aware Configuration As introduced in our complexity analysis, CSPN with large kernel size and long time propagation is time consuming.  ... 
arXiv:1911.05377v2 fatcat:dbuq6nvr7jh6fb2dnh5r5af46m

Fast Sliding Window Classification with Convolutional Neural Networks

Henry G. R. Gouk, Anthony M. Blake
2014 Proceedings of the 29th International Conference on Image and Vision Computing New Zealand - IVCNZ '14  
Unfortunately, due to the high model complexity CNNs often cannot be used for object detection tasks with real-time constraints, where many predictions have to be made on sub-windows of a large input image  ...  Convolutional Neural Networks (CNNs) have repeatedly been shown to be the state of the art method for natural signal classification -image classification in particular.  ...  Once again, the solution is rather simple; only convert convolutional layers with sufficiently large kernels to use the frequency domain forward propagation method.  ... 
doi:10.1145/2683405.2683429 dblp:conf/ivcnz/GoukB14 fatcat:junfq22ybfaazb7xraxszxkl4u

CSPN++: Learning Context and Resource Aware Convolutional Spatial Propagation Networks for Depth Completion

Xinjing Cheng, Peng Wang, Chenye Guan, Ruigang Yang
2020 PROCEEDINGS OF THE THIRTIETH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE AND THE TWENTY-EIGHTH INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE CONFERENCE  
In this paper, we propose CSPN++, which further improves its effectiveness and efficiency by learning adaptive convolutional kernel sizes and the number of iterations for the propagation, thus the context  ...  In our experiments, we find weighted assembling can lead to significant accuracy improvements, which we referred to as "context-aware CSPN", while weighted selection, "resource-aware CSPN" can reduce the  ...  Resource Aware Configuration As introduced in our complexity analysis, CSPN with large kernel size and long time propagation is time consuming.  ... 
doi:10.1609/aaai.v34i07.6635 fatcat:rt5jzhgjzvd2hiu74zf4axeumy

Argus CNN Accelerator Based on Kernel Clustering and Resource-Aware Pruning

Damjan M. Rakanovic, Vuk Vranjkovic, Rastislav J. R. Struharik
2021 Elektronika ir Elektrotechnika  
The proposed CNN pruning algorithm first combines similar kernels into clusters, which are then pruned using the same regular pruning pattern.  ...  Paper proposes a two-step Convolutional Neural Network (CNN) pruning algorithm and resource-efficient Field-programmable gate array (FPGA) CNN accelerator named "Argus".  ...  This approach is very efficient when the layers are large in terms of the number of kernels and IFMs, like in VGG- 16 and AlexNet.  ... 
doi:10.5755/j02.eie.28922 fatcat:no3wm6g2obhlvjk65ebj7rrppq

Depth-Aware CNN for RGB-D Segmentation [chapter]

Weiyue Wang, Ulrich Neumann
2018 Lecture Notes in Computer Science  
Convolutional neural networks (CNN) are limited by the lack of capability to handle geometric information due to the fixed grid kernel structure.  ...  To address these issues, we present Depth-aware CNN by introducing two intuitive, flexible and effective operations: depth-aware convolution and depth-aware average pooling.  ...  The convolution layer parameters are denoted as "C[kernel size]-[number of channels]-[dilation]". "DC" and "Davgpool" represent depth-aware convolution and depth-aware average pooling respectively.  ... 
doi:10.1007/978-3-030-01252-6_9 fatcat:ecvunnbzqvb3xj7xk7sbmaxgmu

Fully Automated 3D Segmentation of MR-Imaged Calf Muscle Compartments: Neighborhood Relationship Enhanced Fully Convolutional Network [article]

Zhihui Guo, Honghai Zhang, Zhi Chen, Ellen van der Plas, Laurie Gutmann, Daniel Thedens, Peggy Nopoulos, Milan Sonka
2020 arXiv   pre-print
We present a novel fully convolutional network (FCN), called FilterNet, that utilizes contextual information in a large neighborhood and embeds edge-aware constraints for individual calf muscle compartment  ...  Achieving clinically acceptable results is a challenging task due to large variations in muscle shape and MR appearance.  ...  However, a large kernel size significantly increases the number of parameters.  ... 
arXiv:2006.11930v2 fatcat:vne3si6ivfeu3gm3av7wyphsgy

Real-time Scene Segmentation Using a Light Deep Neural Network Architecture for Autonomous Robot Navigation on Construction Sites [article]

Khashayar Asadi, Pengyu Chen, Kevin Han, Tianfu Wu, Edgar Lobaton
2019 arXiv   pre-print
., context-awareness, control, localization, and mapping) on an embedded platform.  ...  To overcome this challenge, this paper presents a light and efficient deep neural network architecture to run on an embedded platform in real-time.  ...  Combining the SLAM and Context-Awareness Modules into the same Jetson TX1 would help to mitigate this problem, but the large size of the segmentation model and the high memory usage, make this solution  ... 
arXiv:1901.08630v1 fatcat:czsd3c3fgvgiflxs23mlt6hkge

Self-supervised Exposure Trajectory Recovery for Dynamic Blur Estimation [article]

Youjian Zhang, Chaoyue Wang, Stephen J. Maybank, Dacheng Tao
2020 arXiv   pre-print
Finally, we demonstrate that the estimated exposure trajectories can fit real-world dynamic blurs and further contribute to motion-aware image deblurring and warping-based video extraction from a single  ...  Compare to the vanilla DMPHN(1-2-4), our motion-aware deblurring network can be easily derived by replacing the selected convolutional layers with the proposed motion-aware convolution.  ...  Adding the motion-aware module can already achieve comparable results, and our model largely reduces the memory cost.  ... 
arXiv:2010.02484v1 fatcat:6wmtnmbotffmrle4a2evpzog4a

Multi-scale Location-Aware Kernel Representation for Object Detection

Hao Wang, Qilong Wang, Mingqi Gao, Peihua Li, Wangmeng Zuo
2018 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition  
To this end, we propose a novel Multi-scale Location-aware Kernel Representation (MLKP) to capture high-order statistics of deep features in proposals.  ...  Our M-LKP can be efficiently computed on a modified multi-scale feature map using a low-dimensional polynomial kernel approximation.  ...  Large 1 10 100 Small Med.  ... 
doi:10.1109/cvpr.2018.00136 dblp:conf/cvpr/WangWGLZ18 fatcat:4s7mbdxpezcadn5kih4v2vsl7u

CARAFE++: Unified Content-Aware ReAssembly of FEatures [article]

Jiaqi Wang, Kai Chen, Rui Xu, Ziwei Liu, Chen Change Loy, Dahua Lin
2020 arXiv   pre-print
receptive field. (2) Instead of using a fixed kernel for all samples (e.g. convolution and deconvolution), CARAFE++ generates adaptive kernels on-the-fly to enable instance-specific content-aware handling  ...  CARAFE++ has several appealing properties: (1) Unlike conventional methods such as pooling and interpolation that only exploit sub-pixel neighborhood, CARAFE++ aggregates contextual information within a large  ...  Second, it comes with heavy computational workload when a large kernel size is used.  ... 
arXiv:2012.04733v1 fatcat:qyic2r2krvavxl3qxkvqwhbuzy

3D Neighborhood Convolution: Learning Depth-Aware Features for RGB-D and RGB Semantic Segmentation [article]

Yunlu Chen, Thomas Mensink, Efstratios Gavves
2019 arXiv   pre-print
We introduce 3D Neighborhood Convolution (3DN-Conv), a convolutional operator around 3D neighborhoods.  ...  We propose to define convolutions local with respect to the corresponding point in the 3D real-world space ($x, y, z$), where the depth channel is used to adapt the receptive field of the convolution,  ...  We observe the following: First both depth-aware convolution methods improve over the RGB segmentation baseline by a large margin, as was also noted in [34] .  ... 
arXiv:1910.01460v1 fatcat:dvanvznz4jdfhlpsz7rxmeqzni

Depth-aware CNN for RGB-D Segmentation [article]

Weiyue Wang, Ulrich Neumann
2018 arXiv   pre-print
Convolutional neural networks (CNN) are limited by the lack of capability to handle geometric information due to the fixed grid kernel structure.  ...  To address these issues, we present Depth-aware CNN by introducing two intuitive, flexible and effective operations: depth-aware convolution and depth-aware average pooling.  ...  The convolution layer parameters are denoted as "C[kernel size]-[number of channels]-[dilation]". "DC" and "Davgpool" represent depth-aware convolution and depth-aware average pooling respectively.  ... 
arXiv:1803.06791v1 fatcat:dxogeqi4grc6ngmxiwpvmbjk5a
« Previous Showing results 1 — 15 out of 20,828 results