Convolutional neural network simplification via feature map pruning

Zou, Junhua; Rui, Ting; Zhou, You; Yang, Cheng‐Hong; Zhang, Sai

doi:10.1016/j.compeleceng.2018.01.036

Cited by 41 publications

(23 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The experimental results show that a network layer can be severely pruned and yet continue to be run in the forward direction with coherent and effective activations in a later layer. Our approach also does not re-train after pruning, unlike many previous work in the space [7], [8], [17], [18]. Therefore this research shows that, while later layers are directly impacted by the complete removal of early features, the removal of up to 50% of these early features does not cause a catastrophic collapse of activation strength in later layers.…”

Section: Discussionmentioning

confidence: 76%

“…Li et al uses the absolute magnitude of weights within each feature map as the pruning metric, with low magnitude maps being removed [17]. An alternate approach is to perform Linear Discriminant Analysis, to preserve the class structure in the final layer while reducing the number of feature maps in earlier layers [8]. A recent work, which is currently under review, uses structured sparsity regularization and Stochastic Gradient Descent to prune unnecessary feature maps [18].…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Filter Early, Match Late: Improving Network-Based Visual Place Recognition

Hausler

Jacobson

Milford

2019

2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

View full text Add to dashboard Cite

CNNs have excelled at performing place recognition over time, particularly when the neural network is optimized for localization in the current environmental conditions. In this paper we investigate the concept of feature map filtering, where, rather than using all the activations within a convolutional tensor, only the most useful activations are used. Since specific feature maps encode different visual features, the objective is to remove feature maps that are detract from the ability to recognize a location across appearance changes.Our key innovation is to filter the feature maps in an early convolutional layer, but then continue to run the network and extract a feature vector using a later layer in the same network. By filtering early visual features and extracting a feature vector from a higher, more viewpoint invariant later layer, we demonstrate improved condition and viewpoint invariance. Our approach requires image pairs for training from the deployment environment, but we show that state-of-the-art performance can regularly be achieved with as little as a single training image pair. An exhaustive experimental analysis is performed to determine the full scope of causality between early layer filtering and late layer extraction. For validity, we use three datasets: Oxford RobotCar, Nordland, and Gardens Point, achieving overall superior performance to NetVLAD. The work provides a number of new avenues for exploring CNN optimizations, without full re-training.

show abstract

Section: Discussionmentioning

confidence: 76%

Section: Related Workmentioning

confidence: 99%

Filter Early, Match Late: Improving Network-Based Visual Place Recognition

Hausler

Jacobson

Milford

2019

2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

View full text Add to dashboard Cite

show abstract

“…is research uses VGG-16 for feature extraction. After visualizing the convolutional layer through feature map [38], the feature map of each channel can be obtained, and each channel is fused according to 1 : 1 to obtain the fused feature map. Figure 6 shows the convolutional layer map after channel fusion.…”

Section: Traditional Shallow Expression Featuresmentioning

confidence: 99%

Facial Expression Recognition Algorithm Based on Fusion of Transformed Multilevel Features and Improved Weighted Voting SVM

Meng

Yuan

et al. 2021

Mathematical Problems in Engineering

View full text Add to dashboard Cite

In allusion to the shortcomings of traditional facial expression recognition (FER) that only uses a single feature and the recognition rate is not high, a FER method based on fusion of transformed multilevel features and improved weighted voting SVM (FTMS) is proposed. The algorithm combines the transformed traditional shallow features and convolutional neural network (CNN) deep semantic features and uses an improved weighted voting method to make a comprehensive decision on the results of the four trained SVM classifiers to obtain the final recognition result. The shallow features include local Gabor features, LBP features, and joint geometric features designed in this study, which are composed of distance and deformation characteristics. The deep feature of CNN is the multilayer feature fusion of CNN proposed in this study. This study also proposes to use a better performance SVM classifier with CNN to replace Softmax since the poor distinction between facial expressions. Experiments on the FERPlus database show that the recognition rate of this method is 17.2% higher than that of the traditional CNN, which proves the effectiveness of the fusion of the multilayer convolutional layer features and SVM. FTMS-based facial expression recognition experiments are carried out on the JAFFE and CK+ datasets. Experimental results show that, compared with the single feature, the proposed algorithm has higher recognition rate and robustness and makes full use of the advantages and characteristics of different features.

show abstract

“…How many convolution kernels are there in each convolution layer will get the characteristics of how many channels. After visualizing the convolution layer through feature map [12], we can get the characteristic map of each channel and fuse each channel according to 1:1 to get the fused characteristic map, as shown in Figure 2. Because there are many convolution layers in Inception v3, there are also many convolution kernels in each layer, that is, many channels.…”

Section: Figure 1 Tensorboard Visualization Nodementioning

confidence: 99%

Facial Expression Analysis Based on Fusion Multi-Layer Convolutional Layer Feature Neural Network

Meng¹,

Yuan²,

Yan³

2020

Fuzzy Systems and Data Mining VI

View full text Add to dashboard Cite

Concerning the problem that the current facial expression analysis based on convolutional neural network (CNN) only uses the features of the last convolutional layer but the recognition rate is not high, this paper proposes the use of sub-deep convolutional layer features and builds a CNN model which fuses the features of multi-layer convolutional layers. The model uses a CNN for feature extraction and saves the deepest feature vectors and sub-deep feature vectors of the expression images. The sub-deep feature vector is used as the input of the multilayer CNN established in this paper. The processed fourth convolution layer feature is fused with the deepest feature previously saved to perform facial expression analysis. Experiments are performed on FERPLUS dataset, Cohn-Kanade dataset (CK+) and JAFFE dataset. The experimental results show that the improved network structure proposed in this paper can capture richer feature information during facial expression analysis, which greatly improves the accuracy of expression recognition and the stability of the network. Compared with the original CNN-based facial expression analysis using only the last layer of convolution layer features, using multi-layer fusion features on three kinds of datasets can improve the expression recognition rate by 33.3%, 2.3% and 22%, respectively.

show abstract

Convolutional neural network simplification via feature map pruning

Cited by 41 publications

References 8 publications

Filter Early, Match Late: Improving Network-Based Visual Place Recognition

Filter Early, Match Late: Improving Network-Based Visual Place Recognition

Facial Expression Recognition Algorithm Based on Fusion of Transformed Multilevel Features and Improved Weighted Voting SVM

Facial Expression Analysis Based on Fusion Multi-Layer Convolutional Layer Feature Neural Network

Contact Info

Product

Resources

About