Skip to content

add bilinear upsample layer #1136

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.

Sign up for GitHub

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

Closed
wants to merge 6 commits into from
Closed

add bilinear upsample layer #1136

Changes from 1 commit
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
734afc5
add bilinear upsample layer
Apr 21, 2020
2d0bd18
add doctest. add include.
kczimm Apr 23, 2020
d2fee6d
remove batch/channel loop. add @inbounds and @views.
kczimm Apr 23, 2020
3110531
fixed doc sentence.
kczimm Apr 23, 2020
0d9b84f
add tests
kczimm Apr 23, 2020
a15f02e
fix grammer in doc
kczimm Apr 23, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
41 changes: 41 additions & 0 deletions src/layers/upsample.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,41 @@
struct BilinearUpsample{T<:Integer}
factor::Tuple{T,T}
end

function (b::BilinearUpsample)(x::AbstractArray)
Copy link
Contributor

Choose a reason for hiding this comment

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

Suggested change
function (b::BilinearUpsample)(x::AbstractArray)
function (b::BilinearUpsample)(x::AbstractArray{T, 4}) where T

W, H, C, N = size(x)
Copy link
Contributor

Choose a reason for hiding this comment

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

It might be better to swap W and H but it's still okay to keep it as it is, since most people confuse them.

Copy link
Author

Choose a reason for hiding this comment

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

The data in Flux is stored in WHCN order isn’t it?

Flux.jl/src/layers/conv.jl

Line 17 in 7a32a70

Data should be stored in WHCN order (width, height, # channels, batch size).

Copy link
Contributor

@johnnychen94 johnnychen94 Apr 24, 2020 •
edited
Loading

Choose a reason for hiding this comment

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

It is, but I prefer to read it as a misunderstanding. I just wanted to mention it here in case you're not aware of it.

It's okay to abuse the usage of WH since it's relative to the column/row-first order.


newW = W * b.factor[1]
newH = H * b.factor[2]

out = similar(x, (newW, newH, C, N))

for n = 1:N, c = 1:C, w = 1:newW, h = 1:newH
Copy link
Contributor

Choose a reason for hiding this comment

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

This scalar loop would be extremely slow when x is CuArray

Copy link
Author

Choose a reason for hiding this comment

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

Agreed, but I must admit I don't have experience optimizing Julia for CUDA usage. Do you have suggestions in this case?

Copy link

Choose a reason for hiding this comment

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

Broadcast would be fast on GPU

w₀ = (w - 0.5) / b.factor[1] + 0.5
h₀ = (h - 0.5) / b.factor[2] + 0.5

w1 = floor(Int, w₀)
w2 = w1 + 1
h1 = floor(Int, h₀)
h2 = h1 + 1

i1 = clamp(w1, 1, W)
i2 = clamp(w2, 1, W)
j1 = clamp(h1, 1, H)
j2 = clamp(h2, 1, H)

out[w, h, c, n] =
(
x[i1, j1, c, n] * (w2 - w₀) * (h2 - h₀) +
x[i1, j2, c, n] * (w2 - w₀) * (h₀ - h1) +
x[i2, j1, c, n] * (w₀ - w1) * (h2 - h₀) +
x[i2, j2, c, n] * (w₀ - w1) * (h₀ - h1)
) / (w2 - w1 * h2 - h1)
end

out
end

function Base.show(io::IO, b::BilinearUpsample)
print(io, "BilinearUpsample(", b.factor[1], ", ", b.factor[2], ")")
end