Sementic segmentation
- A classification peoblem for each pixel
- Not about classifications of instances.
Fully Convolutional Networks(FCN)[Long et al., CVPR 2015]
- the first end-to-end architecture for semantic segmentation
- Convolution networks without a flatten layer(input resolution free)
- 1x1 convolution : the same operation with fc layer for each spatial point.
- low resolution of feature map : large receptive -> upsample and apply
- upsampling
- transposed convolution has overlap issues.
- Intrgrates activations from lower layers into prediction
- near input : Fine Low-level Detail Local
- near output : Coarse Semantic Holistic Global
Hypercolumns for object segmentation[Hariharan et al., CVPR 2015]
- integrates activation from lower layers into perdiction
U-Net[Ronneberger et al., MICCAI 2015]
- Contracting path and Expanding path
- Contracting path
- convolution layer and ReLU activation
- maxpooling(compressing spatial dimension and expanding channel dimension)
-Expanding path
- convolution transposed layer(compressing channel dimension and expanding spatial dimension)
- kernel size = 2, stride = 2 : to avoid checker board artifact
DeepLab[Chen et al., ICLR 2015]
- CRF : post-processes a segmentation map to be refined to follow image boundaries.
- Dilated convolution
- large receptive field
- Depthwise separable convolution (proposed by Howard et al.)
- DeepLab v3+[Chen et al., ECCV 2018]
- Atrous spatial pyramid pooling
- Decoder module and upsample
###
피어세션
https://www.notion.so/vgg11-Batch_norm-2280d43fa25e45aaa40c15c035e1f893
'딥러닝 머신러닝 데이터 분석 > BoostCampAITech' 카테고리의 다른 글
[CV] Further topics of segmentation (0) | 2021.09.19 |
---|---|
[CV] Object detection (0) | 2021.09.12 |
[CV] Image Classification 2 (0) | 2021.09.08 |
[CV] 0. Computer Vision OT & 1. Image Classification 1 & 2. Annotation data efficient learning (0) | 2021.09.07 |
[P-Stage] 마스크 데이터 분류 대회 리포트 - 2 (0) | 2021.09.03 |
댓글