DenseASPP for Semantic Segmentation in Street Scenes

By Maoke Yang, Kun Yu, Chi Zhang, Zhiwei Li, Kuiyuan Yang. 2018

1,473 citations recorded by the source corpus.

Semantic image segmentation is a basic street scene understanding task in autonomous driving, where each pixel in a high resolution image is categorized into a set of semantic labels. Unlike other scenarios, objects in autonomous driving scene exhibit very large scale changes, which poses great challenges for high-level feature representation in a sense that multi-scale information must be correctly encoded. To remedy this problem, atrous convolution[14]was introduced to generate features with larger receptive fields without sacrificing spatial resolution. Built upon atrous convolution, Atrous Spatial Pyramid Pooling (ASPP)[2] was proposed to concatenate multiple atrous-convolved features using different dilation rates into a final feature representation. Although ASPP is able to generate multi-scale features, we argue the feature resolution in the scale-axis is not dense enough for the autonomous driving scenario. To this end, we propose Densely connected Atrous Spatial Pyramid Pooling (DenseASPP), which connects a set of atrous convolutional layers in a dense way, such that it generates multi-scale features that not only cover a larger scale range, but also cover that scale range densely, without significantly increasing the model size. We evaluate DenseASPP on the street scene benchmark Cityscapes[4] and achieve state-of-the-art performance.

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

CCNet: Criss-Cross Attention for Semantic Segmentation (2018, methodology): “DenseASPP [34] brought dense connections into ASPP to generate features with various scale.”
Dual Attention Network for Scene Segmentation (2018, methodology): “Moreover, it also surpasses DenseASPP [25], which use more powerful pretrained models than ours.”
Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers (2020, background): “However, it also raises a fundamental limitation that learning long-range dependency information, critical for semantic segmentation in unconstrained scene images [1,48], becomes challenging due to still limited receptive fields.”
Object-Contextual Representations for Semantic Segmentation (2019, background): “The recent works [22,68,83,71] propose various extensions, e.”
SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers (2021, background): “The limited receptive ﬁeld in CNN requires one to resort to context modules such as ASPP [18] that enlarge the receptive ﬁeld but inevitably become heavy.”
Strip Pooling: Rethinking Spatial Pooling for Scene Parsing (2020, methodology): “Other methods for long-range context modeling include: dilated convolutions [5, 8, 6, 57] that aim to widen the receptive fields of CNNs without introducing extra parameters; or global/pyramid pooling [26, 65, 19, 5, 8, 54] that summarizes global clues of the images.”
SPGNet: Semantic Prediction Guidance for Scene Parsing (2019, methodology): “On Cityscapes test set [14], our proposed SPGNet out-performs the strong baseline DenseASPP [66] when only exploiting the ‘ﬁne’ annotations.”
OCNet: Object Context Network for Scene Parsing (2018, methodology): “• Fine-tuning: We adopt the ﬁnetuning strategy proposed by DeepLabv3 [3] and DenseASPP [28] to ﬁne-tune the model with the ﬁne-labeled dataset for extra epochs and further boost the performance to 81 .”
Image Segmentation Using Deep Learning: A Survey (2020, background): “include the DeepLab family [76], multiscale context aggregation [77], Dense Upsampling Convolution and Hybrid Dilated Convolution (DUC-HDC) [78], densely connected Atrous Spatial Pyramid Pooling (DenseASPP) [79], and the”
Foreground-Aware Relation Network for Geospatial Object Segmentation in High Spatial Resolution Remote Sensing Imagery (2020, methodology): “We compared FarSeg with several CNN-based methods from classical to state-of-the-art, including U-Net [38], FCN-8s [33], DenseASPP [44], Deeplab v3 [6], Semantic FPN [32], Deeplab v3+ [8], RefineNet [29], PSPNet [48].”
GFF: Gated Fully Fusion for Semantic Segmentation (2019, methodology): “In addition, contextual information in large receptive ﬁeld is also very important for semantic segmentation as proved by PSPNet (Zhao et al. 2017), ASPP (Chen et al. 2018a) and DenseASPP (Yang et al. 2018).”
Scene Segmentation With Dual Relation-Aware Attention Network (2020, methodology): “Moreover, it also surpasses DenseASPP [49] that uses more powerful pretrained models.”
U-NetPlus: A Modified Encoder-Decoder U-Net Architecture for Semantic and Instance Segmentation of Surgical Instruments from Laparoscopic Images (2019, background): “edu datasets [8], [25] have achieved ground-breaking performance by the virtue of deep convolutional neural networks (CNNs), image segmentation in clinical settings still requires more accuracy and precision, with even minimal segmentation errors being unacceptable.”
Med3D: Transfer Learning for 3D Medical Image Analysis (2019, methodology): “Finally, we concatenate the pre-trained Med3D encoder with DenseASPP [35] network and demonstrate the state-of-the-art performance on the liver segmentation task using a single model.”
SegNeXt: Rethinking Convolutional Attention Design for Semantic Segmentation (2022, methodology): “Method Backbone mIoU (%) DenseASPP [83] ResNet50 57.”