Conv2d in PyTorch

Buy Me a Coffee☕

*Memos:

• My post explains Convolutional Layer.
• My post explains Conv1d().
• My post explains Conv3d().
• My post explains manual_seed().
• My post explains requires_grad.

Conv2d() can get the 3D or 4D tensor of the zero or more elements computed by 2D convolution from the 3D or 4D tensor of zero or more elements as shown below:

*Memos:

• The 1st argument for initialization is in_channels(Required-Type:float). *It must be 0 <= x.
• The 2nd argument for initialization is out_channels(Required-Type:float). *It must be 1 <= x.
• The 3rd argument for initialization is kernel_size(Required-Type:int or tuple or list of int). *It must be 1 <= x.
• The 4th argument for initialization is stride(Optional-Default:1-Type:int or tuple or list of int). *It must be 1 <= x.
• The 5th argument for initialization is padding(Optional-Default:0-Type:int, str or tuple or list of int): *Memos:
  • It must be 0 <= x if not str.
  • It must be either 'valid' or 'same' for str.
• The 6th argument for initialization is dilation(Optional-Default:1-Type:int, tuple or list of int). *It must be 1 <= x.
• The 7th argument for initialization is groups(Optional-Default:1-Type:int). *It must be 1 <= x.
• The 8th argument for initialization is bias(Optional-Default:True-Type:bool). *My post explains bias argument.
• The 9th argument for initialization is padding_mode(Optional-Default:'zeros'-Type:str). *'zeros', 'reflect', 'replicate' or 'circular' can be selected.
• The 10th argument for initialization is device(Optional-Default:None-Type:str, int or device()): *Memos:
  • If it's None, get_default_device() is used. *My post explains get_default_device() and set_default_device().
  • device= can be omitted.
  • My post explains device argument.
• The 11th argument for initialization is dtype(Optional-Default:None-Type:dtype): *Memos:
  • If it's None, get_default_dtype() is used. *My post explains get_default_dtype() and set_default_dtype().
  • dtype= can be omitted.
  • My post explains dtype argument.
• The 1st argument is input(Required-Type:tensor of float or complex): *Memos:
  • It must be the 3D or 4D tensor of zero or more elements.
  • The number of the elements of the 3rd deepest dimension must be same as in_channels.
  • Its device and dtype must be same as Conv2d()'s.
  • complex must be set to dtype of Conv2d() to use a complex tensor.
  • The tensor's requires_grad which is False by default is set to True by Conv2d().
• conv2d1.device and conv2d1.dtype don't work.

import torch
from torch import nn

tensor1 = torch.tensor([[[8., -3., 0., 1., 5., -2.]]])

tensor1.requires_grad
# False

torch.manual_seed(42)

conv2d1 = nn.Conv2d(in_channels=1, out_channels=3, kernel_size=1)
tensor2 = conv2d1(input=tensor1)
tensor2
# tensor([[[7.0349, -1.3750, 0.9186, 1.6831, 4.7413, -0.6105]],
#         [[6.4210, -2.7091, -0.2191, 0.6109, 3.9309, -1.8791]],
#         [[-1.6724, 0.9046, 0.2018, -0.0325, -0.9696, 0.6703]]],
#        grad_fn=<SqueezeBackward1>)

tensor2.requires_grad
# True

conv2d1
# Conv2d(1, 3, kernel_size=(1, 1), stride=(1, 1))

conv2d1.in_channels
# 1

conv2d1.out_channels
# 3

conv2d1.kernel_size
# (1, 1)

conv2d1.stride
# (1, 1)

conv2d1.padding
# (0, 0)

conv2d1.dilation
# (1, 1)

conv2d1.groups
# 1

conv2d1.bias
# Parameter containing:
# tensor([0.9186, -0.2191, 0.2018], requires_grad=True)

conv2d1.padding_mode
# 'zeros'

conv2d1.weight
# Parameter containing:
# tensor([[[[0.7645]]], [[[0.8300]]], [[[-0.2343]]]], requires_grad=True)

torch.manual_seed(42)

conv2d2 = nn.Conv2d(in_channels=3, out_channels=3, kernel_size=1)
conv2d2(input=tensor2)
# tensor([[[5.9849, -2.4511, -0.1504, 0.6165, 3.6841, -1.6842]],
#         [[3.2258, 0.2207, 1.0403, 1.3134, 2.4062, 0.4939]],
#         [[-0.5434, 0.0364, -0.1217, -0.1744, -0.3853, -0.0163]]],
#        grad_fn=<SqueezeBackward1>)
torch.manual_seed(42)

conv2d = nn.Conv2d(in_channels=1, out_channels=3, kernel_size=1, stride=1, 
                   padding=0, dilation=1, groups=1, bias=True,
                   padding_mode='zeros', device=None, dtype=None)
conv2d(input=tensor1)
# tensor([[[7.0349, -1.3750, 0.9186, 1.6831, 4.7413, -0.6105]],
#         [[6.4210, -2.7091, -0.2191, 0.6109, 3.9309, -1.8791]],
#         [[-1.6724, 0.9046, 0.2018, -0.0325, -0.9696, 0.6703]]],
#        grad_fn=<SqueezeBackward1>)

my_tensor = torch.tensor([[[8., -3., 0.],
                           [1., 5., -2.]]])
torch.manual_seed(42)

conv2d = nn.Conv2d(in_channels=1, out_channels=3, kernel_size=1)
conv2d(input=my_tensor)
# tensor([[[7.0349, -1.3750, 0.9186], [1.6831, 4.7413, -0.6105]],
#         [[6.4210, -2.7091, -0.2191], [0.6109, 3.9309, -1.8791]],
#         [[-1.6724, 0.9046, 0.2018], [-0.0325, -0.9696, 0.6703]]],
#        grad_fn=<SqueezeBackward1>)

my_tensor = torch.tensor([[[8.], [-3.], [0.],
                           [1.], [5.], [-2.]]])
torch.manual_seed(42)

conv2d = nn.Conv2d(in_channels=1, out_channels=3, kernel_size=1)
conv2d(input=my_tensor)
# tensor([[[7.0349], [-1.3750], [0.9186], [1.6831], [4.7413], [-0.6105]],
#         [[6.4210], [-2.7091], [-0.2191], [0.6109], [3.9309], [-1.8791]],
#         [[-1.6724], [0.9046], [0.2018], [-0.0325], [-0.9696], [0.6703]]],
# grad_fn=<SqueezeBackward1>)

my_tensor = torch.tensor([[[[8.], [-3.], [0.]],
                           [[1.], [5.], [-2.]]]])
torch.manual_seed(42)

conv2d = nn.Conv2d(in_channels=2, out_channels=3, kernel_size=1)
conv2d(input=my_tensor)
# tensor([[[[4.5675], [0.9684], [-1.5181]],
#          [[-0.2604], [4.1600], [-0.8838]],
#          [[-0.4734], [1.8016], [0.3380]]]],
#        grad_fn=<ConvolutionBackward0>)

my_tensor = torch.tensor([[[[8.+0.j], [-3.+0.j], [0.+0.j]],
                           [[1.+0.j], [5.+0.j], [-2.+0.j]]]])
torch.manual_seed(42)

conv2d = nn.Conv2d(in_channels=2, out_channels=3, kernel_size=1, 
                   dtype=torch.complex64)
conv2d(input=my_tensor)
# tensor([[[[4.6816+5.4406j], [-1.9277+1.5828j], [0.8537-1.2033j]],
#          [[-1.2427+1.4569j], [-0.9155+1.5485j], [1.0295-0.9304j]],
#          [[6.1465-3.9132j], [1.7481+2.3225j], [-0.6841-0.1602j]]]],
#        grad_fn=<AddBackward0>)