Low rank approximation is an effective method in deep neural network (DNN) compression. In view of the fact that the redundancy information content of different network layers is different, a novel iterative low-rank approximation method based on the redundancy of each network layer is proposed. By giving priority to the network layer with higher redundancy, the loss of intrinsic information in each network layer is expected to be reduced and the performance of the compressed model is improved. Experimental results show that the performance of compressed model obtained by this method is improved with a slight reduction in compression ratio. It can be concluded that the proposed method can better retain intrinsic information in the pre-training network.
Image super-resolution methods based on forward-feed convolutional neural networks (CNN) reconstruct the image with more details and sharper texture. However, most of these methods do not consider the influence of high level semantic feature to improve image perceptual effect. In this paper, we propose a deep CNN architecture jointing low-high level feature for image super-resolution. Our method uses 17 weight layers to predict residual between the high resolution and low resolution image. And we joint the low level and high level image features to constraint the network parameters updating. Experimental results validate that our method reconstruct the high resolution images with clear edge and less warp.
Recently, discriminative object trackers based on deep learning have demonstrated excellent performance. However, the tracking accuracy is facing a challenge due to contaminated training samples and different complex scenarios. For this reason, we propose a tracker based on sparse robust samples and convolutional residual learning with multi-feature fusion (SR_MFCRL). First, a sparse robust sample set (SRSS) is introduced to improve robustness of the network. In this process, we first employ sparse representation to estimate the best candidate and then utilize joint detection with response peak value and occlusion detection to determine the contamination degree of the sample. Second, a multifeature fusion residual network (MRN) is proposed and its two base branches to capture response output of different features in order to achieve higher positioning accuracy. Extensive experimental results conducted on OTB-2013 illustrate that the proposed tracker achieves outstanding performance in terms of tracking accuracy and robustness.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.