Tensorflow and pytorch middle note feature map size adjustment to achieve up sampling

1.tensorflow:

tf.image.resize_images(fea, size, method)

'''
fea: feature map
size:  The goal is size, example ：(300, 300)
method = 0  Bilinear interpolation 
method = 1  Nearest neighbor law 
method = 2  Bicubic interpolation 
method = 3  Area interpolation 
'''

Commonly used in , adjustment feature map The size of the , Further add convolution to adjust the number of channels to complete up sampling

2.torch：

torch.nn.functional.interpolate
(x,
size,
scale_factor,
mode="bilinear",
align_corners=True,
recompute_scale_factor=None
)

'''
x: feature map
size:  The goal is size, example (128,128)
scale_factor:  Multiple of change 
mode: 'nearest','linear','bilinear','bicubic','trilinear','area'. Default ：'nearest'
'''

This paper records Basic application mode , About align_corners and recompute_scale_factor For detailed explanation of parameters and some precautions, please refer to :https://blog.csdn.net/qq_50001789/article/details/120297401

Commonly used in , adjustment feature map The size of the , Further add convolution to adjust the number of channels to complete up sampling

Except for this On the sampling The way , The more common up sampling methods are deconvolution

tf2: layers.Conv2DTranspose()

torch: torch.nn.ConvTranspose2d()