语义分割--Attention to Scale: Scale-aware Semantic Image Segmentation
2017-06-08 13:45
1871 查看
Attention to Scale: Scale-aware Semantic Image Segmentation
CVPR2016
http://liangchiehchen.com/projects/DeepLab.html
针对语义分割问题,嵌入多尺度信息是很有必要的,这里我们提出用一个 attention mechanism 来学习每个像素位置的 softly weight the multi-scale features
![](https://oscdn.geek-share.com/Uploads/Images/Content/202008/20/c9453a2c73c204fe4821b5c046b2943a)
attention model 学习对于不同尺度的物体赋予不同的权重
对于提取多尺度特征,目前主要有两种网络结构:
![](https://oscdn.geek-share.com/Uploads/Images/Content/202008/20/15a3574787a5114f6ba40cdddd581e98)
Skip-net 和 Share-net,这里我们认为 Share-net 能够与 attention model 更好的结合,所以采用了 attention model
怎么融合多尺度特征信息了?
![](https://oscdn.geek-share.com/Uploads/Images/Content/202008/20/8cea2f41e94eeff1e2743ed0e1ee794a)
这里我们首先得到权重,再根据权重来融合多尺度特征信息
PASCAL-Person-Part validation set
![](https://oscdn.geek-share.com/Uploads/Images/Content/202008/20/2e2a3ebb9afed489d471950cf7453a2b)
E-Supv: extra supervision The ground truths are downsampled properly during training
![](https://oscdn.geek-share.com/Uploads/Images/Content/202008/20/0e4b6ff5ab509bd149826b917f1242ee)
max-pooling 和 attention model 效果对比:
![](https://oscdn.geek-share.com/Uploads/Images/Content/202008/20/40c8d5dfec488ef8223d9ae0cf9b5b80)
![](https://oscdn.geek-share.com/Uploads/Images/Content/202008/20/f1dd5f564e87c6c58a1ceca983b59843)
PASCAL VOC 2012 test set
![](https://oscdn.geek-share.com/Uploads/Images/Content/202008/20/f9b08948fbd0eb2397692945486d1c25)
PASCAL VOC 2012 validation set
![](https://oscdn.geek-share.com/Uploads/Images/Content/202008/20/a3aaa129f6efdbf338b50d3cc7f459a9)
MS-COCO validation set
CVPR2016
http://liangchiehchen.com/projects/DeepLab.html
针对语义分割问题,嵌入多尺度信息是很有必要的,这里我们提出用一个 attention mechanism 来学习每个像素位置的 softly weight the multi-scale features
attention model 学习对于不同尺度的物体赋予不同的权重
对于提取多尺度特征,目前主要有两种网络结构:
Skip-net 和 Share-net,这里我们认为 Share-net 能够与 attention model 更好的结合,所以采用了 attention model
怎么融合多尺度特征信息了?
这里我们首先得到权重,再根据权重来融合多尺度特征信息
PASCAL-Person-Part validation set
E-Supv: extra supervision The ground truths are downsampled properly during training
max-pooling 和 attention model 效果对比:
PASCAL VOC 2012 test set
PASCAL VOC 2012 validation set
MS-COCO validation set
相关文章推荐
- Attention to Scale: Scale-Aware Semantic Image Segmentation论文阅读笔记
- [Paper Reading] Attention to Scale: Scale-aware Semantic Image Segmentation
- 语义分割--Not All Pixels Are Equal:Difficulty-Aware Semantic Segmentation via Deep Layer Cascade
- DeepLab:深度卷积网络,多孔卷积 和全连接条件随机场 的图像语义分割 Semantic Image Segmentation with Deep Convolutional Nets, Atro
- 语义分割--Learning Object Interactions and Descriptions for Semantic Image Segmentation
- 语义分割 DeepLabv3--Rethinking Atrous Convolution for Semantic Image Segmentation
- 语义分割DeepLab v2--DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolut
- 语义分割--End-to-End Instance Segmentation with Recurrent Attention
- 语义分割--Deep Dual Learning for Semantic Image Segmentation
- 语义分割--Label Refinement Network for Coarse-to-Fine Semantic Segmentation
- 语义分割--Loss Max-Pooling for Semantic Image Segmentation
- 今日技点 | 乱弹 语义分割(Semantic Segmentation)
- 图像分割“Not All Pixels Are Equal: Difficulty-Aware Semantic Segmentation via Deep Layer Cascade”
- 图像分割“Instance-aware Semantic Segmentation via Multi-task Network Cascades”
- 语义分割--Understand Convolution for Semantic Segmentation
- 语义分割--Global Deconvolutional Networks for Semantic Segmentation
- 语义分割 - Semantic Segmentation Papers
- 图像分割“Fully Convolutional Instance-aware Semantic Segmentation”
- CNN for Semantic Segmentation(语义分割,论文,代码,数据集,标注工具,blog)
- 语义分割--Mix-and-Match Tuning for Self-Supervised Semantic Segmentation