Filters








12,348 Hits in 4.3 sec

Refinement of matching costs for stereo disparities using recurrent neural networks

Alper Emlek, Murat Peker
2021 EURASIP Journal on Image and Video Processing  
Exploiting this sequential information, we aimed to determine the position of the correct matching point by using recurrent neural networks, as in the case of speech processing problems.  ...  Žbontar and LeCun proposed MC-CNN [16] in which two CNN-based Siamese networks are introduced named as fast (MC-CNN-Fast) and accurate (MC-CNN-Acrt) networks.  ...  16.03 16.80 17.91 20.14 7.4006 35.12 MC-CNN-Acrt 13.18 13.81 14.76 16.86 6.5936 35.13 MC-CNN-Fast + CR-RNN C 13.53 14.36 15.65 18.66 4.8488 26.84 MC-CNN-Fast + CR-RNN P 12.89  ... 
doi:10.1186/s13640-021-00551-9 fatcat:2xbvb5cjufbjdpibbizhkhb6tq

Speeding up Context-based Sentence Representation Learning with Non-autoregressive Convolutional Decoding [article]

Shuai Tang, Hailin Jin, Chen Fang, Zhaowen Wang, Virginia R. de Sa
2018 arXiv   pre-print
After that, we designed a model which still keeps an RNN as the encoder, while using a non-autoregressive convolutional decoder.  ...  We empirically show that our model is simple and fast while producing rich sentence representations that excel in downstream tasks.  ...  As presented, our designed asymmetric RNN-CNN model has strong transferability, and is overall better than existing unsupervised models in terms of fast training speed and good performance on evaluation  ... 
arXiv:1710.10380v3 fatcat:wgckw2cz7bdb7atat74gg4fcxa

Semantic Regularisation for Recurrent Image Annotation [article]

Feng Liu, Tao Xiang, Timothy M. Hospedales, Wankou Yang, Changyin Sun
2016 arXiv   pre-print
Importantly this makes the end-to-end training of the CNN and RNN slow and ineffective due to the difficulty of back propagating gradients through the RNN to train the CNN.  ...  Existing models use the weakly semantic CNN hidden layer or its transform as the image embedding that provides the interface between the CNN and RNN.  ...  It thus allows for better and more efficient fine-tuning of the CNN module, as well as fast convergence in end-to-end training of the full CNN-RNN model.  ... 
arXiv:1611.05490v1 fatcat:hedw7uovwjgftp2wwi4uvaoi7y

Semantic Regularisation for Recurrent Image Annotation

Feng Liu, Tao Xiang, Timothy M. Hospedales, Wankou Yang, Changyin Sun
2017 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)  
Importantly this makes the end-to-end training of the CNN and RNN slow and ineffective due to the difficulty of back propagating gradients through the RNN to train the CNN.  ...  Existing models use the weakly semantic CNN hidden layer or its transform as the image embedding that provides the interface between the CNN and RNN.  ...  It thus allows for better and more efficient fine-tuning of the CNN module, as well as fast convergence in end-to-end training of the full CNN-RNN model.  ... 
doi:10.1109/cvpr.2017.443 dblp:conf/cvpr/LiuXHYS17 fatcat:gkgkysdyrfdrzjmvyeuzw3xkvu

Speeding up Context-based Sentence Representation Learning with Non-autoregressive Convolutional Decoding

Shuai Tang, Hailin Jin, Chen Fang, Zhaowen Wang, Virginia de Sa
2018 Proceedings of The Third Workshop on Representation Learning for NLP  
After that, we designed a model which still keeps an RNN as the encoder, while using a non-autoregressive convolutional decoder.  ...  We empirically show that our model is simple and fast while producing rich sentence representations that excel in downstream tasks.  ...  As presented, our designed asymmetric RNN-CNN model has strong transferability, and is overall better than existing unsupervised models in terms of fast training speed and good performance on evaluation  ... 
doi:10.18653/v1/w18-3009 dblp:conf/rep4nlp/TangJFWS18 fatcat:azhyp36ptnd7lh7jnzrpn3wtqm

Real-time Memory Efficient Large-pose Face Alignment via Deep Evolutionary Network [article]

Bin Sun, Ming Shao, Siyu Xia, Yun Fu
2019 arXiv   pre-print
Afterward, a simple and effective CNN feature is extracted and fed to Recurrent Neural Network (RNN) for evolutionary learning.  ...  However, impact factors such as large pose variation and computational inefficiency, still hinder its broad implementation.  ...  Note that the experiments of fast RNN module is implemented based on Fast DHM CNN ×2.  ... 
arXiv:1910.11818v2 fatcat:vs4nz7vvfrezpcu2gjjvykwjry

Deep Recurrent Neural Network Reveals a Hierarchy of Process Memory during Dynamic Natural Vision [article]

Junxing Shi, Haiguang Wen, Yizhen Zhang, Kuan Han, Zhongming Liu
2017 bioRxiv   pre-print
Unlike the CNN, the RNN learned spatiotemporal features from videos to enable action recognition.  ...  The extended model, or the recurrent neural network (RNN), embodied a hierarchical and distributed model of process memory as an integral part of visual processing.  ...  As described below, the same training methods were used regardless of whether the RNN or the CNN was used as the feature model.  ... 
doi:10.1101/177196 fatcat:42o7v3pbg5cpbnsh66keatv3py

Exploiting Multi-layer Features Using a CNN-RNN Approach for RGB-D Object Recognition [chapter]

Ali Caglayan, Ahmet Burak Can
2019 Lecture Notes in Computer Science  
It first employs a pre-trained CNN model as the underlying feature extractor to get visual features at different layers for RGB and depth modalities.  ...  In order to utilize the CNN model trained on large-scale RGB datasets for depth domain, depth images are converted to a representation similar to RGB images.  ...  We use the pre-trained VGG-f model as a feature extractor without finetuning. Therefore, the procedure requires no training at feature extraction stage and works fast.  ... 
doi:10.1007/978-3-030-11015-4_51 fatcat:a4iezbiv3vfrraal333akeipba

Deep Learning Based Energy Disaggregation and On/Off Detection of Household Appliances [article]

Jie Jiang, Qiuqiang Kong, Mark Plumbley, Nigel Gilbert
2019 arXiv   pre-print
Based on the same dataset, we show that for the task of on/off detection the second framework, i.e., directly training a binary classifier, achieves better performance in terms of F1 score.  ...  Neural network models can learn complex patterns from large amounts of data and have been shown to outperform the traditional machine learning methods such as variants of hidden Markov models.  ...  ACKNOWLEDGEMENT This work was carried out as part of the "HomeSense: digital sensors for social research" project funded by the Economic and Social Research Council (grant ES/N011589/1) through the National  ... 
arXiv:1908.00941v2 fatcat:fxrwhhdtvjconalzvf4pmncss4

Fast Prospective Detection of Contrast Inflow in X-ray Angiograms with Convolutional Neural Network and Recurrent Neural Network [chapter]

Hua Ma, Pierre Ambrosini, Theo van Walsum
2017 Lecture Notes in Computer Science  
The first approach trains a convolutional neural network (CNN) to distinguish whether a frame has contrast agent or not.  ...  As the proposed methods work in prospective settings and run fast, they have the potential of being used in clinical practice.  ...  While the CNN-based method ran very fast and used on average only 14 ms to process one frame.  ... 
doi:10.1007/978-3-319-66179-7_52 fatcat:n55fyy7g6nbptcbb2kmtgdqvaq

Fast Detection of Distributed Denial of Service Attacks in VoIP Networks Using Convolutional Neural Networks

Waleed Nazih, Yasser Hifny, Wail S. Elkilani, Tamer Mostafa
2020 International Journal of Intelligent Computing and Information Sciences  
Our experiments find that the CNN model achieved a high F1 score (99-100%) as another deep learning approach that utilizes Recurrent Neural Network (RNN) but with less detection time.  ...  Full deep neural network model based on CNN 132 W. Nazih et al  ...  As shown in the table, CNN-AVG, CNN-MAX, and RNN-GUR almost detect different types of DDoS attacks, while the l 1 -SVM didn't detect low-rate attacks.  ... 
doi:10.21608/ijicis.2021.51555.1046 fatcat:dexh6czbk5c3xnmc2xqgyrnarq

Real-Time Object Recognition using Region based Convolution Neural Network and Recursive Neural Network

2019 International journal of recent technology and engineering  
Specifically, we applied the transfer learning approach to pass the load parameters which were pre-trained on the Image web dataset to the RNN portion and follow a custom loss feature for the model to  ...  In particular, a hybrid approach using R-CNN and RNN has been proposed that improve the accuracy of object recognition and learn structured image attributes and begin image analysis.  ...  RNN can be viewed as one of the most efficient systems that find combined working stages of CNN and highway networks.  ... 
doi:10.35940/ijrte.d8326.118419 fatcat:3znfgj5mhrh4xoq6bexwul2hv4

Speech Command Recognition using Artificial Neural Networks

Sushan Poudel, Dr. R Anuradha
2020 JOIV: International Journal on Informatics Visualization  
the RNN as our models.  ...  RNN diagram result showed that among models trained on 20 different speech commands, CNN in combination with the RNN achieved 94.79% validation accuracy, 96.66 test accuracy and 0.117 loss whereas, CNN  ... 
doi:10.30630/joiv.4.2.358 fatcat:pozzuovuqjf6xnrsu3ib63lvya

A Study on a Joint Deep Learning Model for Myanmar Text Classification

Myat Sapal Phyu, Khin Thandar Nwet
2020 2020 IEEE Conference on Computer Applications(ICCA)  
The comparative analysis is performed on the baseline Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN) and their combined model CNN-RNN.  ...  This paper uses pre-trained word vectors to handle the resource-demanding problem and studies the effectiveness of a joint Convolutional Neural Network and Long Short Term Memory (CNN-LSTM) for Myanmar  ...  We greatly thanks all of the researchers who shared pre-trained words vectors publicly and their works very helpful to accomplish our works and very useful for resource-scarce languages.  ... 
doi:10.1109/icca49400.2020.9022809 fatcat:hrbkpumwlbax5jflqbxetmy4bq

Evaluation of Adversarial Training on Different Types of Neural Networks in Deep Learning-based IDSs [article]

Rana Abou Khamis, Ashraf Matrawy
2020 arXiv   pre-print
networks (CNN) and recurrent neural networks (RNN).  ...  We use the min-max approach to formulate the problem of training robust IDS against adversarial examples using two benchmark datasets.  ...  We built ten adversarially trained models for each ANN, CNN, and RNN architectures, as shown in Fig 6 . In total, we built 30 adversarially trained IDS models on top of the six baseline IDS models.  ... 
arXiv:2007.04472v1 fatcat:3qothuaw6fh55idtksh453nd3a
« Previous Showing results 1 — 15 out of 12,348 results