hi,

I have a 1024*1024 dimension feature vector to reduce dimension to 1,1024, so how can I get the final dimension using conv2d operation over 1024*1024. please explain the forward function for this or any other way

I don’t know which dimensions are referenced by the shape `[1024, 1024]`

, but assume the spatial ones.

If so, you could use a conv layer with `kernel_size=(1024, 1)`

to create an output of `[batch_size, out_channels, 1, 1024]`

:

```
conv = nn.Conv2d(3, 4, (1024, 1))
x = torch.randn(2, 3, 1024, 1024)
out = conv(x)
print(out.shape)
# > torch.Size([2, 4, 1, 1024])
```

can you please explain why we use` x = torch.randn(2, 3, 1024, 1024)`

instead of `x = torch.randn(1024,1024)`

`nn.Conv2d`

expects an input in the shape `[batch_size, channels, height, width]`

as described in the docs, so I’ve used random values for the missing dimensions.

ok. in my case its like `x = torch.randn(1, 1, 1024, 1024)`