Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix UNet implementation with arbitrary channel sizes (#243) #276

Open
wants to merge 10 commits into
base: master
Choose a base branch
from
52 changes: 43 additions & 9 deletions src/convnets/unet.jl
Original file line number Diff line number Diff line change
Expand Up @@ -71,18 +71,44 @@ Backbone of any Metalhead ResNet-like model can be used as encoder
- `final`: final block as described in original paper
- `fdownscale`: downscale factor
"""
function unet(encoder_backbone, imgdims, outplanes::Integer,
final::Any = unet_final_block, fdownscale::Integer = 0)
backbonelayers = collect(flatten_chains(encoder_backbone))
layers = unetlayers(backbonelayers, imgdims; m_middle = unet_middle_block,
skip_upscale = fdownscale)
function unet(encoder_backbone, imgdims, inchannels::Integer,outplanes::Integer,
final::Any = unet_final_block, fdownscale::Integer = 0)
backbonelayers = collect(flatten_chains(encoder_backbone))
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

please pay attention to the formatting, you lost the indentation here


outsz = Flux.outputsize(layers, imgdims)
layers = Chain(layers, final(outsz[end - 1], outplanes))
# Adjusting input size to include channels
adjusted_imgdims = (imgdims..., inchannels, 1)

return layers
layers = unetlayers(backbonelayers, adjusted_imgdims; m_middle = unet_middle_block,
skip_upscale = fdownscale)

outsz = Flux.outputsize(layers, adjusted_imgdims)
layers = Chain(layers, final(outsz[end - 1], outplanes))

return layers
end
function modify_first_conv_layer_advanced(encoder_backbone, inchannels)
layers = [layer for layer in encoder_backbone.layers] # Create a mutable array from the layers
modified = false
for index in 1:length(layers)
if isa(layers[index], Flux.Conv) && !modified
layer = layers[index]
outchannels = size(layer.weight, 1) # The number of output channels
kernel_size = (size(layer.weight, 3), size(layer.weight, 4)) # Kernel size
stride = layer.stride
pad = layer.pad


new_conv_layer = Flux.Conv(kernel_size, inchannels => outchannels, stride=stride, pad=pad)
layers[index] = new_conv_layer # Replace the old layer with the new one

modified = true # Mark as modified to avoid changing any other Conv layer
end
end
return Flux.Chain(layers...) # Reconstruct the model with the modified layers
end



"""
UNet(imsize::Dims{2} = (256, 256), inchannels::Integer = 3, outplanes::Integer = 3,
encoder_backbone = Metalhead.backbone(DenseNet(121)); pretrain::Bool = false)
Expand Down Expand Up @@ -110,17 +136,25 @@ See also [`Metalhead.unet`](@ref).
struct UNet
layers::Any
end
@functor UNet

@functor UNet
function UNet(imsize::Dims{2} = (256, 256), inchannels::Integer = 3, outplanes::Integer = 3,
encoder_backbone = Metalhead.backbone(DenseNet(121)); pretrain::Bool = false)

if inchannels != 3
encoder_backbone = modify_first_conv_layer_advanced(encoder_backbone, inchannels)

end

layers = unet(encoder_backbone, (imsize..., inchannels, 1), outplanes)
model = UNet(layers)
if pretrain

artifact_name = "UNet"
loadpretrain!(model, artifact_name)
end
return model
end


(m::UNet)(x::AbstractArray) = m.layers(x)
Loading