上采样方法大PK（Upsample，Interpolate，resize，Transposed convolution，deconv，Unpool，Pixelshuffle）

刺骨的言语ヽ痛彻心扉 2023-03-01 10:46 1阅读 0赞

**目录**

Upsample:

Interpolate，resize:

Transposed convolution，deconv:

Unpool：

Pixelshuffle：

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

# **Upsample:** #

**Pytorch example:**

import torch
    import torch.nn as nn
    import torch.nn.functional as F
    
    input = torch.arange(1, 5).view(1, 1, 2, 2).float()
    print(input)
    #tensor([[[[1., 2.],
    #          [3., 4.]]]])
    upsample = nn.Upsample(scale_factor=2, mode='nearest')
    print(upsample(input))
    #tensor([[[[1., 1., 2., 2.],
    #          [1., 1., 2., 2.],
    #          [3., 3., 4., 4.],
    #          [3., 3., 4., 4.]]]])
    upsample = nn.Upsample(scale_factor=2, mode='bilinear')
    print(upsample(input))
    #tensor([[[[1.0000, 1.2500, 1.7500, 2.0000],
    #          [1.5000, 1.7500, 2.2500, 2.5000],
    #          [2.5000, 2.7500, 3.2500, 3.5000],
    #          [3.0000, 3.2500, 3.7500, 4.0000]]]])

**tensorflow example:**

import tensorflow as tf
    
    x = tf.random_normal((1, 4,4,9))
    print(x.get_shape().as_list()) #[1, 4, 4, 9]
    y = tf.keras.layers.UpSampling2D(size=(1, 2))(x)
    print(y.get_shape().as_list())#[1, 4, 8, 9]

# **Interpolate，resize:** #

**Pytorch example:**

import torch
    import torch.nn as nn
    import torch.nn.functional as F
    
    input = torch.arange(1, 5).view(1, 1, 2, 2).float()
    print(input)
    #tensor([[[[1., 2.],
    #          [3., 4.]]]])
    upsample = F.interpolate(input, size=(4, 4), mode='nearest')
    print(upsample)
    #tensor([[[[1., 1., 2., 2.],
    #          [1., 1., 2., 2.],
    #          [3., 3., 4., 4.],
    #          [3., 3., 4., 4.]]]])
    upsample = F.interpolate(input, size=(4, 4), mode='bilinear')
    print(upsample)
    #tensor([[[[1.0000, 1.2500, 1.7500, 2.0000],
    #          [1.5000, 1.7500, 2.2500, 2.5000],
    #          [2.5000, 2.7500, 3.2500, 3.5000],
    #          [3.0000, 3.2500, 3.7500, 4.0000]]]])

**tensorflow example:**

import tensorflow as tf
    image = tf.constant([
     [1,0,0,0,0],
     [0,1,0,0,0],
     [0,0,1,0,0],
     [0,0,0,1,0],
     [0,0,0,0,1],
    ])
    
    image = image[tf.newaxis, ..., tf.newaxis]# Add "batch" and "channels" dimensions
    print(image.shape.as_list())  # [1, 5, 5, 1]
    resized_image = tf.image.resize_images(image, [3,5]) #tf1.9.0
    print(resized_image.get_shape().as_list())#[1, 3, 5, 1]

# **Transposed convolution，deconv:** #

出自论文，[Learning Deconvolution Network for Semantic Segmentation][]

![20200724113403951.png][]

**Pytorch example:**

import torch
    import torch.nn as nn
    import torch.nn.functional as F
    
    input = torch.randn(1, 3, 6, 6)
    upsample = nn.ConvTranspose2d(3, 16, 3, stride=2, padding=1)
    output = upsample(input, output_size=(1,16,12,12))
    print(output.size()) #torch.Size([1, 16, 12, 12])

**tensorflow example:**

import tensorflow as tf
    import tensorflow.contrib.slim as slim
    
    input = tf.random_normal((1, 4,4,9))
    print(input.get_shape().as_list()) #[1, 4, 4, 9]
    
    pred = slim.conv2d_transpose(input, 16, kernel_size=[3, 3], stride=2)# tf 1.15.3
    print(pred.get_shape().as_list())#[1, 8, 8, 16]

# **Unpool：** #

出自论文，

[Visualizing and Understanding Convolutional Networks][]

![20200724113515414.png][]

[SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation][SegNet_ A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation]

![20200724113548709.png][]

**Pytorch example:**

import torch
    from torch import nn
     
    input = torch.Tensor([1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16]).view(1,1,4,4)
    print(input)
    downsample ,index= nn.MaxPool2d(kernel_size=2, stride=2,return_indices=True)(input)
    out = nn.MaxUnpool2d(kernel_size =2, stride=2)(downsample,index)
    print(out)
     
    “””
    tensor([[[[ 1.,  2.,  3.,  4.],
              [ 5.,  6.,  7.,  8.],
              [ 9., 10., 11., 12.],
              [13., 14., 15., 16.]]]])
    tensor([[[[ 0.,  0.,  0.,  0.],
              [ 0.,  6.,  0.,  8.],
              [ 0.,  0.,  0.,  0.],
              [ 0., 14.,  0., 16.]]]])
    “””

# **Pixelshuffle：** #

出自论文：[Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network][]

![20200724113706636.png][]

**Pytorch example:**

import torch
    import torch.nn as nn
    
    pixel_shuffle = nn.PixelShuffle(3)
    input = torch.randn(1, 9, 4, 4)
    output = pixel_shuffle(input)
    print(output.size()) #torch.Size([1, 1, 12, 12])

**tensorflow example:**

import tensorflow as tf
    x = tf.random_normal((1, 4,4,9))
    print(x.shape) #(1, 4, 4, 9)
    y = tf.depth_to_space(x,3)
    with tf.Session() as sess:
        z = sess.run(y)
        print(z.shape) #(1, 12, 12, 1)

[Learning Deconvolution Network for Semantic Segmentation]: https://arxiv.org/pdf/1505.04366.pdf
[20200724113403951.png]: /images/20230208/6d0deb70c72240cabb32ab7d602fb086.png
[Visualizing and Understanding Convolutional Networks]: https://arxiv.org/pdf/1311.2901v3.pdf
[20200724113515414.png]: /images/20230208/8811a8787b964598a85033a6626cffd1.png
[SegNet_ A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation]: https://arxiv.org/pdf/1511.00561v1.pdf
[20200724113548709.png]: /images/20230208/509114085bf644db86fb567ca57c79b0.png
[Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network]: https://arxiv.org/abs/1609.05158
[20200724113706636.png]: https://img-blog.csdnimg.cn/20200724113706636.png