论文阅读:《Associative Embedding:End-to-End Learning for Joint Detection and Grouping》
2017-10-23 19:58
627 查看
概述
论文提出了一种single-stage,end-to-end的关节点检测和分组方法,这不同于以往的multi-stage的关节点检测方法,在MPII和COCO数据集上达到新的state-of-the-art的效果,超越最近的Mask RCNN和Google GMI。从人体姿态估计方法上属于bottom-up的方法,即先检测关节点,再对关节点进行分组。在COCO测试集上mAP达到0.655。尽管论文声称的结果是很好的,但是rejected by ICCV 2017。
关节点检测
关节点检测使用的ECCV 2016的单人姿态估计的网络stacked hourglass,在其上做了一些修改,在每一次下采样时增加输出通道的个数,同时individual layers的残差模型改为3*3的卷积结构,其他结构不变。关节点分组
使用stacked hourglass网络的输出是对每一个pixel预测detection score,从单人到多人的姿态估计就是将heatmap由单个点的激活,到多个点的激活,从而检测出多个关节点。引入一个新的“tag”heatmap,用于对每一个pixel打上标签,标签相近的关节点认为是属于同一个人的,这样完成对关节点的分类。如果有m个关节点,就会有2m个通道,其中m个用于检测,m个用于分组标签。
Loss
Prediction loss+grouping loss,其中prediction loss和之前的hourglass相同;grouping loss由两部分组成,一是单个人的K个关节点之间的关联尽可能小,为了保证之后的tag相似,二是多个人之间的关节点平均关联尽可能大,为了区分不同人的tag,grouping loss由两部分组合而成。相关文章推荐
- 姿态检测整理--06-Associative Embedding: End-to-End Learning for Joint Detection and Grouping
- 论文阅读:End-to-End Learning of Deformable Mixture of Parts and Deep Convolutional Neural Networks for H
- 1604.Joint Detection and Identification Feature Learning for Person Search论文阅读笔记
- 论文阅读: End-to-end Learning of Action Detection from Frame Glimpses in Videos
- 论文笔记之:Let there be Color!: Joint End-to-end Learning of Global and Local Image Priors for Automatic
- 论文阅读:Joint Learning of Single-image and Cross-image Representations for Person Re-identification
- 论文阅读:Deep MANTA: A Coarse-to-fine Many-Task Network for joint 2D and 3D vehicle analysis
- 论文笔记之:Let there be Color!: Joint End-to-end Learning of Global and Local Image Priors for Automatic
- 论文阅读(Xiang Bai——【PAMI2017】An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text Recognition)
- 论文笔记之---Joint Detection and Identification Feature Learning for Person Search
- 论文阅读:《LIP: Self-supervised Structure-sensitive Learning and A New Benchmark for Human Parsing》
- 论文阅读——Generic and Scalable Framework for Automated Time-series Anomaly Detection
- 论文阅读:《Neural Machine Translation by Jointly Learning to Align and Translate》
- 论文阅读(3)--SPDA-CNN: Unifying Semantic Part Detection and Abstraction for Fine-grained Recognition
- APPLYING DEEP LEARNING TO ANSWER SELECTION: A STUDY AND AN OPEN TASK 论文阅读
- Deep Region and Multi-label Learning for Facial Action Unit Detection简要论文笔记
- Joint Deep Learning For Pedestrian Detection(论文笔记-深度学习:行人检测)
- End-to-end learning of action detection from frame glimpses in videos 阅读笔记
- Joint Patch and Multi-label Learning for Facial Action Unit Detection
- 论文阅读(Zhuoyao Zhong——【aixiv2016】DeepText A Unified Framework for Text Proposal Generation and Text Detection in Natural Images)