Residual Conv-Deconv Grid Network for Semantic Segmentation

By Damien Fourure, Rémi Emonet, Élisa Fromont, Damien Muselet, Alain Trémeau, Christian Wolf. 2017

238 citations recorded by the source corpus.

This paper presents GridNet, a new Convolutional Neural Network (CNN) architecture for semantic image segmentation (full scene labelling). Classical neural networks are implemented as one stream from the input to the output with subsampling operators applied in the stream in order to reduce the feature maps size and to increase the receptive field for the final prediction. However, for semantic image segmentation, where the task consists in providing a semantic class to each pixel of an image, feature maps reduction is harmful because it leads to a resolution loss in the output prediction. To tackle this problem, our GridNet follows a grid pattern allowing multiple interconnected streams to work at different resolutions. We show that our network generalizes many well known networks such as conv-deconv, residual or U-Net networks. GridNet is trained from scratch and achieves competitive results on the Cityscapes dataset.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

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

In Computer Vision and Pattern Recognition

Citations in the corpus

Deep High-Resolution Representation Learning for Visual Recognition (2019, background): “GridNet [42] is like a combination of multiple U-Nets and includes two symmetric information exchange stages: the first stage passes information only from high resolution to low resolution, and the second stage passes information only from low resolution to high resolution.”
Deep High-Resolution Representation Learning for Human Pose Estimation (2019, background): “There are related multi-scale networks for classification and segmentation [5, 8, 74, 81, 30, 76, 55, 56, 24, 83, 55, 52, 18].”
GridDehazeNet: Attention-Based Multi-Scale Network for Image Dehazing (2019, methodology): “The backbone module is an enhanced version of Grid-Net [7] originally proposed for semantic segmentation.”
Softmax Splatting for Video Frame Interpolation (2020, methodology): “We employ a GridNet [11] architecture with three rows and six columns for this task.”
High-Resolution Representations for Labeling Pixels and Regions (2019, background): “Representative works include GridNet [30], convolutional neural fabrics [86], interlinked CNNs [132], and the recently-developed high-resolution networks (HRNet) [91] that is our interest.”
UNet++: A Nested U-Net Architecture for Medical Image Segmentation (2018, unknown): “The other two recent related works are GridNet [3] and Mask-RCNN [4].”
Context-Aware Synthesis for Video Frame Interpolation (2018, methodology): “We extend the GridNet architecture to generate the final interpolation result from the two pre-warped frames and their context maps [10].”
HRFormer: High-Resolution Transformer for Dense Prediction (2021, methodology): “In the development of high-resolution convolutional neural networks, the community has developed three main paths including: (i) applying dilated convolutions to remove some down-sample layers [6, 51], (ii) recovering high-resolution representations from low-resolution representations with decoders ”
UNet++: Redesigning Skip Connections to Exploit Multiscale Features in Image Segmentation (2019, background): “[45] propose GridNet, which is an encoder-decoder architecture wherein the feature maps are wired in a grid fashion, generalizing several classical segmentation architectures.”
BasicVSR++: Improving Video Super-Resolution with Enhanced Propagation and Alignment (2021, methodology): “Grid-like designs are seen in various vision tasks such as object detection [5, 30, 34], semantic segmentation [7, 30, 34, 43], and frame interpolation [25].”
Portrait shadow manipulation (2020, methodology): “GridNet is a grid-like architecture of rows and columns, where each row is a stream that processes features with resolution kept unchanged, and columns connect the streams by downsampling or upsampling the features.”
Learning Enriched Features for Fast Image Restoration and Enhancement (2022, background): “Therefore, a similar mechanism of collecting multiscale spatial information in the same layer is more effective when incorporated with in CNNs [105], [106], [107], [108].”
Hierarchical Neural Architecture Search for Deep Stereo Matching (2020, background): “Pioneer works [35, 36] propose a super-net that embed a large number of architectures in a grid arrangement and adopt them for the task of semantic segmentation.”
Learning Enriched Features for Real Image Restoration and Enhancement (2020, background): “Therefore, such a mechanism of collecting multi-scale spatial information in the same layer needs to be incorporated in CNNs [46,92,32,93].”