How to apply different kernels for each example in a batch

[ChrisLiu2 (Xujin Liu)]

I’m implementing a dynamic convolution where the kernel is calculated from another network instead of directly learned, but the F.Conv2d function doesn’t seem to support a kernel of size [batch, in, out, H, W]. Is there a workaround for this? Thank you.

[1chimaruGin (Ichimaru Gin)]

torch.nn.functional.conv2d(input, weight, bias=None, stride=1, padding=0, dilation=1, groups=1) → Tensor

filters = torch.randn(8,4,3,3)
inputs = torch.randn(1,4,5,5)
F.conv2d(inputs, filters, padding=1)

[ChrisLiu2 (Xujin Liu)]

Sorry my question was confusing, I edited the question a bit. My question is how to do convolution using different filters for each batch.

[1chimaruGin (Ichimaru Gin)]

Apply different convolutions to a batch of tensors

So, you have filters in a tensor w of shape: [batch_size, out_channels, in_channels, kernel_height, kernel_width], and images in a tensor x of shape: [batch_size, in_channels, in_height, in_width], and you want an output of shape: [batch_size, out_channels, out_height, out_width], where the i_th output is the convolution of the i_th image by the i_th filter ? You can do it by merging the batch and channel dimensions together, and use a grouped convolution, with groups=batch_size : o = to…