ParseNet: Looking Wider to See Better

By Wei David Liu, Andrew Rabinovich, Alexander C. Berg. 2015

1,260 citations recorded by the source corpus.

We present a technique for adding global context to deep convolutional networks for semantic segmentation. The approach is simple, using the average feature for a layer to augment the features at each location. In addition, we study several idiosyncrasies of training, significantly increasing the performance of baseline networks (e.g. from FCN). When we add our proposed global feature, and a technique for learning normalization parameters, accuracy increases consistently even over our improved versions of the baselines. Our proposed approach, ParseNet, achieves state-of-the-art performance on SiftFlow and PASCAL-Context with small additional computational cost over baselines, and near current state-of-the-art performance on PASCAL VOC 2012 semantic segmentation with a simple approach. Code is available at this https URL .

Subfields: Artificial Intelligence, 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 Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Fully convolutional networks for semantic segmentation (2014, neutral): “ParseNet [41] normalizes features for fusion and captures context with global pooling.”
Growing a Brain: Fine-Tuning by Increasing Model Capacity (2017, methodology): “To reconcile the learning pace of the new and pre-existing units, we introduce an additional normalization and adaptive scaling scheme in width augmented networks, which is inspired by the recent work on combining multi-scale pre-trained CNN features from different layers [23].”
CCNet: Criss-Cross Attention for Semantic Segmentation (2018, methodology): “GCN [27] utilized global convolution module and ParseNet [24] utilized global pooling to harvest context information for global representations.”
Feature Pyramid Networks for Object Detection (2016, methodology): “Several other approaches (HyperNet [18], ParseNet [23], and ION [2]) concatenate features of multiple layers before computing predictions, which is equivalent to summing transformed features.”
Fully Convolutional Instance-Aware Semantic Segmentation (2016, background): “The FCNs are extended with global context [28], multi-scale feature fusion [4], and deconvolution [31].”
SSD: Single Shot MultiBox Detector (2015, methodology): “Similarly, [13] showed that adding global context pooled from the topmost feature map can help smooth the segmentation results.”
Beyond Skip Connections: Top-Down Modulation for Object Detection (2016, methodology): “After the resurgence of ConvNets for image classification [8, 29], they have been successfully adopted for a variety of computer vision tasks such as object detection [16, 17, 41, 48], semantic segmentation [3, 6, 34, 35], instance segmentation [21, 22, 38], pose estimation [49, 50], depth estimatio”
The Cityscapes Dataset for Semantic Urban Scene Understanding (2016, background): “Other work takes advantage of deep learning for explicitly integrating global scene context in the prediction 5 of pixel-wise semantic labels, in particular through CNNs [4,39,44,67] or Recurrent Neural Networks (RNNs) [8,52].”
Inside-Outside Net: Detecting Objects in Context with Skip Pooling and Recurrent Neural Networks (2015, methodology): “These skip-layers [34, 23, 24, 12] span multiple spatial resolutions and levels of feature abstraction.”
DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs (2016, background): “lar to [86], we also found that employing a “poly” learning”
Pyramid Scene Parsing Network (2016, methodology): “Detailed per-class results on testing set are shown in Table 8.”
Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation (2018, methodology): “Additionally, DeepLabv3 augments the Atrous Spatial Pyramid Pooling module, which probes convolutional features at multiple scales by applying atrous convolution with different rates, with the image-level features [47].”
Rethinking Atrous Convolution for Semantic Image Segmentation (2017, methodology): “The image-level features are exploited in ParseNet [51] for global context information.”
Learning Representations for Automatic Colorization (2016, methodology): “[22] compensate for such variability by applying layer-wise L2 normalization.”
Path Aggregation Network for Instance Segmentation (2018, methodology): “Global pooling was used in PSPNet [67] and ParseNet [43] to greatly improve quality of semantic segmentation.”
SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation (2015, background): “than many other recent architectures [2], [4], [10], [17].”
Attention to Scale: Scale-Aware Semantic Image Segmentation (2015, methodology): “We consider ParseNet as a special case to exploit multi-scale features, where the whole image is summarized by the image-level feature.”
Large Kernel Matters — Improve Semantic Segmentation by Global Convolutional Network (2017, background): “Among the ﬁrst, Zoom-out [26] proposes a hand-crafted hierarchical context features, while ParseNet [23] adds a global pooling branch to extract context information.”