ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation

By Adam Paszke, Abhishek Chaurasia, Sangpil Kim, Eugenio Culurciello. 2016

2,375 citations recorded by the source corpus.

The ability to perform pixel-wise semantic segmentation in real-time is of paramount importance in mobile applications. Recent deep neural networks aimed at this task have the disadvantage of requiring a large number of floating point operations and have long run-times that hinder their usability. In this paper, we propose a novel deep neural network architecture named ENet (efficient neural network), created specifically for tasks requiring low latency operation. ENet is up to 18$\times$ faster, requires 75$\times$ less FLOPs, has 79$\times$ less parameters, and provides similar or better accuracy to existing models. We have tested it on CamVid, Cityscapes and SUN datasets and report on comparisons with existing state-of-the-art methods, and the trade-offs between accuracy and processing time of a network. We present performance measurements of the proposed architecture on embedded systems and suggest possible software improvements that could make ENet even faster.

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

WoodScape: A Multi-Task, Multi-Camera Fisheye Dataset for Autonomous Driving (2019, methodology): “We use ENet [41] to generate our baseline results.”
AlterSGD: Finding Flat Minima for Continual Learning by Alternative Training (2021, background): “In addition, to improve the segmentation network efﬁciency for applications, light-weight design [33, 44] and network compression [13] can also incorporate and boost the advances for this task.”
Driving Policy Transfer via Modularity and Abstraction (2018, methodology): “We now briefly describe CIL, and refer the reader to Codevilla et al. [8] for further details.”
An Analysis of Deep Neural Network Models for Practical Applications (2016, background): “, 2013), ENet (Paszke et al., 2016) for ImageNet (Culurciello, 2016), GoogLeNet (Szegedy et al.”
YOLACT: Real-Time Instance Segmentation (2019, methodology): “Real-time Instance Segmentation While real-time object detection [30, 34, 35, 36], and semantic segmentation [2, 41, 33, 11, 47] methods exist, few works have focused on real-time instance segmentation.”
Full-Resolution Residual Networks for Semantic Segmentation in Street Scenes (2016, neutral): “Published Unpublished LRR[20] Adelaide[34] Dilation[55] ENet[44] SegNet[2] DeepLab[43] FRRNA/B Dilation[55] LRR[20] FRRNB Figure 6.”
BiSeNet: Bilateral Segmentation Network for Real-time Semantic Segmentation (2018, methodology): “Recently, the algorithms [1,17,25,39] of real-time semantic segmentation have shown that there are mainly three approaches to accelerate the model.”
Not All Pixels Are Equal: Difficulty-Aware Semantic Segmentation via Deep Layer Cascade (2017, methodology): “Another line of research [1, 21, 27] alleviates this problem by using lightweight network architectures.”
Image Segmentation Using Deep Learning: A Survey (2020, background): “Some of most important include the DeepLab family [78], multiscale context aggregation [79], dense upsampling convolution and hybrid dilatedconvolution (DUC-HDC) [80], densely connected Atrous Spatial Pyramid Pooling (DenseASPP) [81], and the efficient neural network (ENet) [82].”
ESPNet: Efficient Spatial Pyramid of Dilated Convolutions for Semantic Segmentation (2018, methodology): “3 Segmentation results on other datasets Unseen dataset: Table 1a compares the performance of ESPNet to that of ENet [20] and ERFNet [21] on an unseen dataset.”
ESPNetv2: A Light-Weight, Power Efficient, and General Purpose Convolutional Neural Network (2018, methodology): “Under the same computational constraints, ESPNetv2 is 4% and 2% more accurate than ENet [32] and ESPNet [28] respectively.”
Efficient ConvNet for real-time semantic segmentation (2017, methodology): “On the other side, some works have focused on efficiency by proposing architectures that can reach real-time segmentation [9][10], but this is normally at the expense of accuracy (e.”
Real-Time Joint Semantic Segmentation and Depth Estimation Using Asymmetric Annotations (2018, background): “and few approaches are currently available [6], [18]–[20].”
TRADI: Tracking deep neural network weight distributions (2019, methodology): “In the second experiment, we train a Enet DNN [43] for semantic segmentation on CamVid dataset [7].”
RangeNet ++: Fast and Accurate LiDAR Semantic Segmentation (2019, methodology): “This spawned the creation of more efficient approaches such as Bonnet [11], ENet [13], ERFNet [17], and Mobilenets V2 [18], which leverage the law of diminishing returns to find the best trade-off between runtime, the number of parameters, and accuracy.”
3D-SIS: 3D Semantic Instance Segmentation of RGB-D Scans (2018, methodology): “To this end, we first pre-train a 2D semantic segmentation network based on the ENet architecture [21].”