└── README.md


/README.md:
--------------------------------------------------------------------------------
 1 | # Toward fast and accurate human pose estimation via soft-gated skip connections
 2 | [![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/toward-fast-and-accurate-human-pose/pose-estimation-on-leeds-sports-poses)](https://paperswithcode.com/sota/pose-estimation-on-leeds-sports-poses?p=toward-fast-and-accurate-human-pose)
 3 | [![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/toward-fast-and-accurate-human-pose/pose-estimation-on-mpii-human-pose)](https://paperswithcode.com/sota/pose-estimation-on-mpii-human-pose?p=toward-fast-and-accurate-human-pose)
 4 | # INTRODUCTION
 5 | Being one of the most challenging computer vision problems with a multitude of applications, human pose estimation
 6 | has been one of the primary research areas that the computer
 7 | vision community tried to solve with Deep Learning and
 8 | Convolutional Neural Networks (CNNs). Given that the
 9 | results produced by existing state-of-the-art methods look
10 | at least impressive both qualitatively and quantitatively, it is
11 | natural to question how much progress can be expected on
12 | this problem over the next years and whether there is room
13 | for further improvement.
14 | 
15 | # Abstract of Research Paper
16 | This paper is on highly accurate and highly
17 | efficient human pose estimation. Recent works based on Fully
18 | Convolutional Networks (FCNs) have demonstrated excellent
19 | results for this difficult problem. While residual connections
20 | within FCNs have proved to be quintessential for achieving
21 | high accuracy, we re-analyze this design choice in the context of
22 | improving both the accuracy and the efficiency over the state-ofthe-art. In particular, we make the following contributions: (a)
23 | We propose gated skip connections with per-channel learnable
24 | parameters to control the data flow for each channel within the
25 | module within the macro-module. (b) We introduce a hybrid
26 | network that combines the HourGlass and U-Net architectures
27 | which minimizes the number of identity connections within the
28 | network and increases the performance for the same parameter
29 | budget. Our model achieves state-of-the-art results on the MPII
30 | and LSP datasets. In addition, with a reduction of 3× in model
31 | size and complexity, we show no decrease in performance when
32 | compared to the original HourGlass network.
33 | # Skip -Gates 
34 | ![img](https://img2020.cnblogs.com/blog/1033571/202009/1033571-20200907190733717-1742101000.png)
35 | 
36 | # Feature Integration
37 | ![imh](https://img2020.cnblogs.com/blog/1033571/202009/1033571-20200907192227591-124027453.png)
38 | 
39 | # References
40 | https://arxiv.org/pdf/2002.11098v1.pdf
41 | 
42 | # Research Paper 
43 | @1adrianb is the original author of the paper 
44 | 


--------------------------------------------------------------------------------