ReNet: A Recurrent Neural Network Based Alternative to Convolutional Networks

By Francesco Visin, Kyle Kastner, Kyunghyun Cho, Matteo Matteucci, Aaron C. Courville, Yoshua Bengio. 2015

282 citations recorded by the source corpus.

In this paper, we propose a deep neural network architecture for object recognition based on recurrent neural networks. The proposed network, called ReNet, replaces the ubiquitous convolution+pooling layer of the deep convolutional neural network with four recurrent neural networks that sweep horizontally and vertically in both directions across the image. We evaluate the proposed ReNet on three widely-used benchmark datasets; MNIST, CIFAR-10 and SVHN. The result suggests that ReNet is a viable alternative to the deep convolutional neural network, and that further investigation is needed.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Model Architectures

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Model Architectures

Citations in the corpus

Inside-Outside Net: Detecting Objects in Context with Skip Pooling and Recurrent Neural Networks (2015, methodology): “While the technical methods employed (spatial RNNs [10, 2, 40], skip-layer connections [34, 23, 24, 12]) have precedents in the literature, we demonstrate that their well-executed combination has an unexpectedly positive impact on the detector’s accuracy.”
Semantic Image Segmentation with Task-Specific Edge Detection Using CNNs and a Discriminatively Trained Domain Transform (2015, methodology): “Some recent work has attempted to model spatial long range dependency with recurrent networks for computer vision tasks [17, 41, 35, 4, 43].”
MelNet: A Generative Model for Audio in the Frequency Domain (2019, methodology): “This approach differs from the tightly coupled multidimensional recurrences used by MDRNNs [17, 16] and Grid LSTMs [25] and more closely resembles the approach taken by ReNet [52].”
Learning Affinity via Spatial Propagation Networks (2017, background): “The one-way connection is a direct extension of sequential recurrent propagation [8, 31, 13].”
Image Segmentation Using Deep Learning: A Survey (2020, methodology): “This model is mainly based on ReNet [83], which was developed for image classification.”
CNN-RNN: a large-scale hierarchical image classification framework (2017, unknown): “Similar work appeared in [46].”
Social LSTM: Human Trajectory Prediction in Crowded Spaces (2016, methodology): “RNN models have also proven to be effective for tasks with densely connected data such as semantic segmentation [76], scene parsing [51] and even as an alternative to Convolutional Neural Networks [65].”
Combining the Best of Convolutional Layers and Recurrent Layers: A Hybrid Network for Semantic Segmentation (2016, methodology): “Following the naming convention in [31], we refer to a spatially recurrent”
Visual Attention Methods in Deep Learning: An In-Depth Survey (2022, background): “To generate global attention, each pixel needs to “see” via ReNet [167] with four recurrent neural networks sweeping horizontally and vertically.”
Spatial–Temporal Recurrent Neural Network for Emotion Recognition (2017, methodology): “For this reason, RNN is usually combined with CNN to model spatio-temporal information in some recognition tasks [22]–[24], [28], [29].”
Survey on semantic segmentation using deep learning techniques (2019, methodology): “The proposed model is a modiﬁed and extended version of ReNet [65].”
ReSeg: A Recurrent Neural Network for Object Segmentation (2015, methodology): “More recently, recurrent neural networks have begun to be employed in a few computation vision tasks (see, e.g., Visin et al., 2015; Kalchbrenner et al., 2015; Chen et al., 2015; Graves & Schmidhuber, 2009).”
High-Resolution Shape Completion Using Deep Neural Networks for Global Structure and Local Geometry Inference (2017, methodology): “Inspired by the LSTM-CF model [16] and ReNet [29], each depth image passes through two convolutional layers, each of which is followed by a max-pooling layer.”
Recent Advances in Recurrent Neural Networks (2017, background): “Furthermore, RNNs can also leverage inherent sequential patterns in image analysis and video processing that are often ignored by other architectures [53].”
Learning Recursive Filters for Low-Level Vision via a Hybrid Neural Network (2016, methodology): “As an important class of neural networks, RNNs have been used for modeling contextual information in sequential data [12,13,14].”
PiCANet: Learning Pixel-Wise Contextual Attention for Saliency Detection (2017, methodology): “Here we employ a more effective and efficient ReNet model [35], which uses four recurrent neural networks to sweep an image both horizontally and vertically along both directions, to incorporate the global context.”