Amplia la resolució d'il·lustracions d'estil anime utilitzant xarxes neuronals convolucionals profundes. I suporta fotos.
+ + +Einzelbild Super-Auflösungs Konvertierung für Bilder im Anime Stil mithilfe eines künstlichen neuronalen Netzwerk. Zusätzlich dazu unterstützt es auch Fotos.
diff --git a/assets/index.es.html b/assets/index.es.html index 5e33cd7..833d939 100644 --- a/assets/index.es.html +++ b/assets/index.es.html @@ -62,6 +62,14 @@ 한국어 + / + + Nederlands + + / + + Català +
Mejora la resolución de ilustraciones de tipo anime utilizando redes neuronales convolucionales. También es compatible con fotografías.
diff --git a/assets/index.fr.html b/assets/index.fr.html index 38eb97d..ee53dfe 100644 --- a/assets/index.fr.html +++ b/assets/index.fr.html @@ -62,6 +62,14 @@ 한국어 + / + + Nederlands + + / + + Català +
Outil permettant la mise à l'échelle de n'importe quelle image de style "anime", par le biais de réseaux de neurones convolutionnels. Fonctionne également sur les photos.
diff --git a/assets/index.html b/assets/index.html index b4f2dd1..97c52eb 100644 --- a/assets/index.html +++ b/assets/index.html @@ -62,6 +62,14 @@ 한국어 + / + + Nederlands + + / + + Català +
Single-Image Super-Resolution for Anime-Style Art using Deep Convolutional Neural Networks. And it supports photo.
diff --git a/assets/index.ja.html b/assets/index.ja.html index 736a535..2a90683 100644 --- a/assets/index.ja.html +++ b/assets/index.ja.html @@ -62,6 +62,14 @@ 한국어 + / + + Nederlands + + / + + Català +
深層畳み込みニューラルネットワークによる二次元画像のための超解像システム。 写真にも対応。
diff --git a/assets/index.ko.html b/assets/index.ko.html index e319857..01922b0 100644 --- a/assets/index.ko.html +++ b/assets/index.ko.html @@ -62,11 +62,19 @@ 한국어 + / + + Nederlands + + / + + Català + -
컨볼루젼 신경망(CNN) 시스템을 통한 딥 러닝으로 2차원 이미지의 해상도를 개선합니다. 그림 뿐만 아니라 사진에도 대응합니다.
+컨볼루션 신경망(CNN) 시스템을 통한 딥 러닝으로 2차원 이미지의 해상도를 개선합니다. 그림뿐만 아니라 사진도 지원합니다.
- 실제 데모 보기
+ 실제 예제 보기
|
@@ -83,7 +91,7 @@
Single-Image Super-Resolutie voor Anime-Style afbeeldingen met behulp van Deep Convolutional Neural Networks. Oh, en het ondersteunt foto's.
+
+ Toon volledige demonstratie
+
+ |
+
+ Ga naar GitHub
+
+ Single-Image Super-Resolution for Anime-Style Art using Deep Convolutional Neural Networks. And it supports photo.
diff --git a/assets/index.ru.html b/assets/index.ru.html
index fbaf4e6..f7c0729 100644
--- a/assets/index.ru.html
+++ b/assets/index.ru.html
@@ -62,6 +62,14 @@
한국어
+ /
+
+ Nederlands
+
+ /
+
+ Català
+
Waifu2x позволяет увеличивать в 2 раза рисованные изображения, например аниме или арт, а также устранять шум на изображении (преимущественно артефакты сжатия JPEG). Теперь также поддерживаются фотографии.
diff --git a/assets/index.tr.html b/assets/index.tr.html
index 2315d8e..9f5199f 100644
--- a/assets/index.tr.html
+++ b/assets/index.tr.html
@@ -62,6 +62,14 @@
한국어
+ /
+
+ Nederlands
+
+ /
+
+ Català
+
Derin Sarmalı Sinirsel Ağ ile çalışan anime stili çalışmalar için süper çözünürlük yükselticidir. Normal fotoğrafları da destekler.
diff --git a/assets/index.zh-CN.html b/assets/index.zh-CN.html
index c7fff74..ac7ad21 100644
--- a/assets/index.zh-CN.html
+++ b/assets/index.zh-CN.html
@@ -62,6 +62,14 @@
한국어
+ /
+
+ Nederlands
+
+ /
+
+ Català
+
本程序使用卷积神经网络对动漫风格的图片进行放大操作（支持照片）。
diff --git a/assets/index.zh-TW.html b/assets/index.zh-TW.html
index 8bef4e5..7698cc0 100644
--- a/assets/index.zh-TW.html
+++ b/assets/index.zh-TW.html
@@ -62,6 +62,14 @@
한국어
+ /
+
+ Nederlands
+
+ /
+
+ Català
+
使用深度卷積神經網路的二次元圖片超解析系統。同時支援照片處理。
diff --git a/images/layer_outputs/layer-0.png b/images/layer_outputs/layer-0.png
new file mode 100644
index 0000000..8246d4a
Binary files /dev/null and b/images/layer_outputs/layer-0.png differ
diff --git a/images/layer_outputs/layer-1.png b/images/layer_outputs/layer-1.png
new file mode 100644
index 0000000..8e3d742
Binary files /dev/null and b/images/layer_outputs/layer-1.png differ
diff --git a/images/layer_outputs/layer-2.png b/images/layer_outputs/layer-2.png
new file mode 100644
index 0000000..742ec30
Binary files /dev/null and b/images/layer_outputs/layer-2.png differ
diff --git a/images/layer_outputs/layer-3.png b/images/layer_outputs/layer-3.png
new file mode 100644
index 0000000..8623bdf
Binary files /dev/null and b/images/layer_outputs/layer-3.png differ
diff --git a/images/layer_outputs/layer-4.png b/images/layer_outputs/layer-4.png
new file mode 100644
index 0000000..60585af
Binary files /dev/null and b/images/layer_outputs/layer-4.png differ
diff --git a/images/layer_outputs/layer-5.png b/images/layer_outputs/layer-5.png
new file mode 100644
index 0000000..1d2032e
Binary files /dev/null and b/images/layer_outputs/layer-5.png differ
diff --git a/images/layer_outputs/layer-6.png b/images/layer_outputs/layer-6.png
new file mode 100644
index 0000000..649eb02
Binary files /dev/null and b/images/layer_outputs/layer-6.png differ
diff --git a/images/layer_outputs/layer-7.png b/images/layer_outputs/layer-7.png
new file mode 100644
index 0000000..8899a0d
Binary files /dev/null and b/images/layer_outputs/layer-7.png differ
diff --git a/lib/compression.lua b/lib/compression.lua
index 6593c82..a31e0ed 100644
--- a/lib/compression.lua
+++ b/lib/compression.lua
@@ -1,5 +1,5 @@
-- snapply compression for ByteTensor
-require 'snappy'
+local snappy = require 'snappy'
local compression = {}
compression.compress = function (bt)
@@ -9,7 +9,7 @@ end
compression.decompress = function(data)
local size = data[1]
local dec = snappy.decompress(data[2]:string())
- local bt = torch.ByteTensor(unpack(torch.totable(size)))
+ local bt = torch.ByteTensor(table.unpack(torch.totable(size)))
bt:storage():string(dec)
return bt
end
diff --git a/lib/image_loader.lua b/lib/image_loader.lua
index f0739a1..2d956f7 100644
--- a/lib/image_loader.lua
+++ b/lib/image_loader.lua
@@ -135,7 +135,7 @@ function image_loader.load_float(file)
if not fp then
error(file .. ": failed to load image")
end
- local buff = fp:read("*a")
+ local buff = fp:read("*all")
fp:close()
return image_loader.decode_float(buff)
end
@@ -144,7 +144,7 @@ function image_loader.load_byte(file)
if not fp then
error(file .. ": failed to load image")
end
- local buff = fp:read("*a")
+ local buff = fp:read("*all")
fp:close()
return image_loader.decode_byte(buff)
end
diff --git a/lib/settings.lua b/lib/settings.lua
index f9161a7..846a0f3 100644
--- a/lib/settings.lua
+++ b/lib/settings.lua
@@ -176,6 +176,10 @@ end
settings.images = string.format("%s/images.t7", settings.data_dir)
settings.image_list = string.format("%s/image_list.txt", settings.data_dir)
+-- patch for lua52
+if not math.log10 then
+ math.log10 = function(x) return math.log(x, 10) end
+end
if settings.gpu:len() > 0 then
local gpus = {}
local gpu_string = utils.split(settings.gpu, ",")
diff --git a/tools/benchmark.lua b/tools/benchmark.lua
index ebc7ae6..3cacddb 100644
--- a/tools/benchmark.lua
+++ b/tools/benchmark.lua
@@ -83,6 +83,11 @@ if opt.output_dir:len() > 0 then
dir.makepath(opt.output_dir)
end
+-- patch for lua52
+if not math.log10 then
+ math.log10 = function(x) return math.log(x, 10) end
+end
+
local function rgb2y_matlab(x)
local y = torch.Tensor(1, x:size(2), x:size(3)):zero()
x = iproc.byte2float(x)
diff --git a/tools/visualize_layer_output.lua b/tools/visualize_layer_output.lua
new file mode 100644
index 0000000..9c7fa2e
--- /dev/null
+++ b/tools/visualize_layer_output.lua
@@ -0,0 +1,138 @@
+require 'pl'
+local __FILE__ = (function() return string.gsub(debug.getinfo(2, 'S').source, "^@", "") end)()
+package.path = path.join(path.dirname(__FILE__), "..", "lib", "?.lua;") .. package.path
+require 'sys'
+require 'w2nn'
+local iproc = require 'iproc'
+local reconstruct = require 'reconstruct'
+local image_loader = require 'image_loader'
+
+local CONV_LAYERS = {"nn.SpatialConvolutionMM",
+ "cudnn.SpatialConvolution",
+ "nn.SpatialFullConvolution",
+ "cudnn.SpatialFullConvolution"
+}
+local ACTIVATION_LAYERS = {"nn.ReLU",
+ "nn.LeakyReLU",
+ "w2nn.LeakyReLU",
+ "cudnn.ReLU",
+ "nn.SoftMax",
+ "cudnn.SoftMax"
+}
+
+local function includes(s, a)
+ for i = 1, #a do
+ if s == a[i] then
+ return true
+ end
+ end
+ return false
+end
+local function count_conv_layers(seq)
+ local count = 0
+ for k = 1, #seq.modules do
+ local mod = seq.modules[k]
+ local name = torch.typename(mod)
+ if name == "nn.ConcatTable" or includes(name, CONV_LAYERS) then
+ count = count + 1
+ end
+ end
+ return count
+end
+local function strip_conv_layers(seq, limit)
+ local new_seq = nn.Sequential()
+ local count = 0
+ for k = 1, #seq.modules do
+ local mod = seq.modules[k]
+ local name = torch.typename(mod)
+ if name == "nn.ConcatTable" or includes(name, CONV_LAYERS) then
+ new_seq:add(mod)
+ count = count + 1
+ if count == limit then
+ if seq.modules[k+1] ~= nil and
+ includes(torch.typename(seq.modules[k+1]), ACTIVATION_LAYERS) then
+ new_seq:add(seq.modules[k+1])
+ end
+ return new_seq
+ end
+ else
+ new_seq:add(mod)
+ end
+ end
+ return new_seq
+end
+local function save_layer_outputs(x, model, out)
+ local count = count_conv_layers(model)
+ print("conv layer count", count)
+ local output_file = path.join(out, string.format("layer-%d.png", 0))
+ image.save(output_file, x)
+ print("* save layer output " .. 0 .. ": " .. output_file)
+ for i = 1, count do
+ output_file = path.join(out, string.format("layer-%d.png", i))
+ print("* save layer output " .. i .. ": " .. output_file)
+ local test_model = strip_conv_layers(model, i)
+ test_model:cuda()
+ test_model:evaluate()
+ local z = test_model:forward(x:reshape(1, x:size(1), x:size(2), x:size(3)):cuda()):float()
+ z = z:reshape(z:size(2), z:size(3), z:size(4)) -- drop batch dim
+ z = image.toDisplayTensor({input=z, padding=2})
+ image.save(output_file, z)
+ collectgarbage()
+ end
+end
+
+local cmd = torch.CmdLine()
+cmd:text()
+cmd:text("waifu2x - visualize layer output")
+cmd:text("Options:")
+cmd:option("-i", "images/miku_small.png", 'path to input image')
+cmd:option("-scale", 2, 'scale factor')
+cmd:option("-o", "./layer_outputs", 'path to output dir')
+cmd:option("-model_dir", "./models/upconv_7/art", 'path to model directory')
+cmd:option("-name", "user", 'model name for user method')
+cmd:option("-m", "noise_scale", 'method (noise|scale|noise_scale|user)')
+cmd:option("-noise_level", 1, '(1|2|3)')
+cmd:option("-force_cudnn", 0, 'use cuDNN backend (0|1)')
+cmd:option("-gpu", 1, 'Device ID')
+
+local opt = cmd:parse(arg)
+cutorch.setDevice(opt.gpu)
+opt.force_cudnn = opt.force_cudnn == 1
+opt.model_path = path.join(opt.model_dir, string.format("%s_model.t7", opt.name))
+
+local x, meta = image_loader.load_float(opt.i)
+if x:size(2) > 256 or x:size(3) > 256 then
+ error(string.format("input image is too large: %dx%d", x:size(3), x:size(2)))
+end
+local model = nil
+local new_x = nil
+
+if opt.m == "noise" then
+ local model_path = path.join(opt.model_dir, ("noise%d_model.t7"):format(opt.noise_level))
+ model = w2nn.load_model(model_path, opt.force_cudnn)
+ if not model then
+ error("Load Error: " .. model_path)
+ end
+elseif opt.m == "scale" then
+ local model_path = path.join(opt.model_dir, ("scale%.1fx_model.t7"):format(opt.scale))
+ model = w2nn.load_model(model_path, opt.force_cudnn)
+ if not model then
+ error("Load Error: " .. model_path)
+ end
+elseif opt.m == "noise_scale" then
+ local model_path = path.join(opt.model_dir, ("noise%d_scale%.1fx_model.t7"):format(opt.noise_level, opt.scale))
+ model = w2nn.load_model(model_path, opt.force_cudnn)
+elseif opt.m == "user" then
+ local model_path = opt.model_path
+ model = w2nn.load_model(model_path, opt.force_cudnn)
+ if not model then
+ error("Load Error: " .. model_path)
+ end
+else
+ error("undefined method:" .. opt.method)
+end
+assert(model ~= nil)
+assert(x ~= nil)
+
+dir.makepath(opt.o)
+save_layer_outputs(x, model, opt.o)
diff --git a/web.lua b/web.lua
index e27c88a..1e58433 100644
--- a/web.lua
+++ b/web.lua
@@ -12,6 +12,7 @@ local iproc = require 'iproc'
local reconstruct = require 'reconstruct'
local image_loader = require 'image_loader'
local alpha_util = require 'alpha_util'
+local compression = require 'compression'
local gm = require 'graphicsmagick'
-- Note: turbo and xlua has different implementation of string:split().
@@ -34,6 +35,7 @@ cmd:option("-curl_request_timeout", 60, "request_timeout for curl")
cmd:option("-curl_connect_timeout", 60, "connect_timeout for curl")
cmd:option("-curl_max_redirects", 2, "max_redirects for curl")
cmd:option("-max_body_size", 5 * 1024 * 1024, "maximum allowed size for uploaded files")
+cmd:option("-cache_max", 200, "number of cached images on RAM")
local opt = cmd:parse(arg)
cutorch.setDevice(opt.gpu)
@@ -75,6 +77,7 @@ local CLEANUP_MODEL = false -- if you are using the low memory GPU, you could us
local CACHE_DIR = path.join(ROOT, "cache")
local MAX_NOISE_IMAGE = opt.max_pixels
local MAX_SCALE_IMAGE = (math.sqrt(opt.max_pixels) / 2)^2
+local PNG_DEPTH = 8
local CURL_OPTIONS = {
request_timeout = opt.curl_request_timeout,
connect_timeout = opt.curl_connect_timeout,
@@ -167,26 +170,73 @@ local function cleanup_model(model)
model:clearState() -- release GPU memory
end
end
-local function convert(x, meta, options)
- local cache_file = path.join(CACHE_DIR, options.prefix .. ".png")
- local alpha_cache_file = path.join(CACHE_DIR, options.alpha_prefix .. ".png")
- local alpha = meta.alpha
- local alpha_orig = alpha
- if path.exists(alpha_cache_file) then
- alpha = image_loader.load_float(alpha_cache_file)
- if alpha:dim() == 2 then
- alpha = alpha:reshape(1, alpha:size(1), alpha:size(2))
- end
- if alpha:size(1) == 3 then
- alpha = image.rgb2y(alpha)
+-- cache
+local g_cache = {}
+local function cache_count()
+ local count = 0
+ for _ in pairs(g_cache) do
+ count = count + 1
+ end
+ return count
+end
+local function cache_remove_old()
+ local old_time = nil
+ local old_key = nil
+ for k, v in pairs(g_cache) do
+ if old_time == nil or old_time > v.updated_at then
+ old_key = k
+ old_time = v.updated_at
end
end
- if path.exists(cache_file) then
- x = image_loader.load_float(cache_file)
+ if old_key then
+ g_cache[old_key] = nil
+ end
+end
+local function cache_compress(raw_image)
+ if raw_image then
+ compressed_image = compression.compress(iproc.float2byte(raw_image))
+ return compressed_image
+ else
+ return nil
+ end
+end
+local function cache_decompress(compressed_image)
+ if compressed_image then
+ local raw_image = compression.decompress(compressed_image)
+ return iproc.byte2float(raw_image)
+ else
+ return nil
+ end
+end
+local function cache_get(filename)
+ local cache = g_cache[filename]
+ if cache then
+ return {image = cache_decompress(cache.image),
+ alpha = cache_decompress(cache.alpha)}
+ else
+ return nil
+ end
+end
+local function cache_put(filename, image, alpha)
+ g_cache[filename] = {image = cache_compress(image),
+ alpha = cache_compress(alpha),
+ updated_at = os.time()};
+ local count = cache_count(g_cache)
+ if count > opt.cache_max then
+ cache_remove_old()
+ end
+end
+local function convert(x, meta, options)
+ local cache_file = path.join(CACHE_DIR, options.prefix .. ".png")
+ local alpha = meta.alpha
+ local alpha_orig = alpha
+ local cache = cache_get(cache_file)
+
+ if cache then
meta = tablex.copy(meta)
- meta.alpha = alpha
- return x, meta
+ meta.alpha = cache.alpha
+ return cache.image, meta
else
local model = nil
if options.style == "art" then
@@ -209,7 +259,6 @@ local function convert(x, meta, options)
if not (alpha:size(2) == x:size(2) and alpha:size(3) == x:size(3)) then
alpha = reconstruct.scale(model.scale, 2.0, alpha,
opt.crop_size, opt.batch_size)
- image_loader.save_png(alpha_cache_file, alpha)
cleanup_model(model.scale)
end
end
@@ -223,7 +272,7 @@ local function convert(x, meta, options)
x, opt.crop_size, opt.batch_size)
cleanup_model(model[options.method])
end
- image_loader.save_png(cache_file, x)
+ cache_put(cache_file, x, alpha)
meta = tablex.copy(meta)
meta.alpha = alpha
return x, meta
@@ -259,6 +308,12 @@ function APIHandler:post()
local style = self:get_argument("style", "art")
local download = (self:get_argument("download", "")):len()
+ if client_disconnected(self) then
+ self:set_status(400)
+ self:write("client disconnected")
+ return
+ end
+
if tta_level == 0 then
tta_level = auto_tta_level(x, scale)
end
@@ -321,7 +376,7 @@ function APIHandler:post()
name = uuid() .. ".png"
end
local blob = image_loader.encode_png(alpha_util.composite(x, meta.alpha),
- tablex.update({depth = 8, inplace = true}, meta))
+ tablex.update({depth = PNG_DEPTH, inplace = true}, meta))
self:set_header("Content-Length", string.format("%d", #blob))
if download > 0 then
@@ -354,6 +409,8 @@ local index_tr = file.read(path.join(ROOT, "assets", "index.tr.html"))
local index_zh_cn = file.read(path.join(ROOT, "assets", "index.zh-CN.html"))
local index_zh_tw = file.read(path.join(ROOT, "assets", "index.zh-TW.html"))
local index_ko = file.read(path.join(ROOT, "assets", "index.ko.html"))
+local index_nl = file.read(path.join(ROOT, "assets", "index.nl.html"))
+local index_ca = file.read(path.join(ROOT, "assets", "index.ca.html"))
local index_en = file.read(path.join(ROOT, "assets", "index.html"))
function FormHandler:get()
local lang = self.request.headers:get("Accept-Language")
@@ -382,6 +439,10 @@ function FormHandler:get()
self:write(index_zh_tw)
elseif langs[1] == "ko" then
self:write(index_ko)
+ elseif langs[1] == "nl" then
+ self:write(index_nl)
+ elseif langs[1] == "ca" or langs[1] == "ca-ES" or langs[1] == "ca-FR" or langs[1] == "ca-IT" or langs[1] == "ca-AD" then
+ self:write(index_ca)
else
self:write(index_en)
end
diff --git a/webgen/locales/ca.yml b/webgen/locales/ca.yml
new file mode 100644
index 0000000..3ae770d
--- /dev/null
+++ b/webgen/locales/ca.yml
@@ -0,0 +1,30 @@
+---
+description: Amplia la resolució d'il·lustracions d'estil anime utilitzant xarxes neuronals convolucionals profundes. I suporta fotos.
+show_demonstration: Mostra'n una demostració
+go_to_github: Vés al GitHub
+image_choosing: Escolliu una imatge
+type_url: Escriviu una URL
+choose_file: O escolliu un fitxer
+file_limits: "Límits: Mida: 5MB, Reducció de soroll: 3000x3000px, Sobreescalat: 1500x1500px."
+style: Estil
+artwork: Il·lustració
+photo: Foto
+noise_reduction: Reducció de soroll
+expect_jpeg: espera artefactes JPEG
+nr_none: Cap
+nr_low: Baixa
+nr_medium: Mitjana
+nr_high: Alta
+nr_highest: Molt alta
+nr_hint: "Utilitzeu la reducció de soroll si la imatge té soroll, si no, podeu aconseguir l'efecte contrari."
+upscaling: Sobreescalat
+up_none: Cap
+tta: TTA
+tta_best_effort: "n-pass auto"
+tta_rule: "Mostra la regla de `n-pass auto`."
+tta_hint: "El TTA estabilitza la qualitat però limita la mida de la imatge d'origen."
+button_convert: Converteix
+button_download: Descarrega
+hints:
+ - "Si utilitzeu el Firefox, premeu la tecla Control+S per desar la imatge. L'opció \"Desa la imatge\" no funciona."
+
diff --git a/webgen/locales/ko.yml b/webgen/locales/ko.yml
index c8e6d06..46befcd 100644
--- a/webgen/locales/ko.yml
+++ b/webgen/locales/ko.yml
@@ -1,24 +1,25 @@
---
-description: 컨볼루젼 신경망(CNN) 시스템을 통한 딥 러닝으로 2차원 이미지의 해상도를 개선합니다. 그림 뿐만 아니라 사진에도 대응합니다.
-show_demonstration: 실제 데모 보기
+description: 컨볼루션 신경망(CNN) 시스템을 통한 딥 러닝으로 2차원 이미지의 해상도를 개선합니다. 그림뿐만 아니라 사진도 지원합니다.
+show_demonstration: 실제 예제 보기
go_to_github: GitHub에서 보기
image_choosing: 이미지 선택
type_url: URL 입력
choose_file: 또는 파일 업로드
-file_limits: "제한: 크기: 5MB, 노이즈 감소: 3000x3000px, 업스케일링: 1500x1500px."
+file_limits: "제한: 크기: 5MB, 노이즈 감소: 3000x3000px, 확대: 1500x1500px."
style: 스타일
artwork: 그림
photo: 사진
noise_reduction: 노이즈 감소
-expect_jpeg: JPEG 아티팩트 제외
+expect_jpeg: JPEG 이미지에 권장
nr_none: 없음
+nr_low: 낮음
nr_medium: 중간
nr_high: 높음
-nr_highest: 최상
-nr_hint: "실제로 노이즈가 있는 이미지에만 사용해야 하며, 그렇지 않을 경우 사진의 세세한 부분이 뭉개질 수 있습니다."
-upscaling: 업스케일링
+nr_highest: 최고
+nr_hint: "실제로 노이즈가 있는 이미지에만 사용해야 하며, 그렇지 않을 경우 세세한 부분이 뭉개질 수 있습니다."
+upscaling: 확대
up_none: 없음
button_convert: 변환
button_download: 다운로드
hints:
- - "FireFox에서 이미지를 저장하려면 Ctrl+S를 누르세요. 오른쪽 클릭 > \"이미지 저장\"을 이용하면 정상적으로 동작하지 않습니다."
+ - "Firefox에서 이미지를 저장하려면 Ctrl+S를 누르십시오. ‘이미지 저장’ 옵션은 작동하지 않습니다."
diff --git a/webgen/locales/nl.yml b/webgen/locales/nl.yml
new file mode 100644
index 0000000..d740242
--- /dev/null
+++ b/webgen/locales/nl.yml
@@ -0,0 +1,30 @@
+---
+description: Single-Image Super-Resolutie voor Anime-Style afbeeldingen met behulp van Deep Convolutional Neural Networks. Oh, en het ondersteunt foto's.
+show_demonstration: Toon volledige demonstratie
+go_to_github: Ga naar GitHub
+image_choosing: Afbeelding selecteren
+type_url: Typ je URL
+choose_file: Of selecteer een bestand
+file_limits: "Limieten: Grootte: 5MB, Ruis vermindering: 3000x3000px, Vergroting: 1500x1500px."
+style: Stijl
+artwork: Artwork
+photo: Foto
+noise_reduction: Ruis vermindering
+expect_jpeg: verwacht JPEG artifact
+nr_none: Geen
+nr_low: Laag
+nr_medium: Gemiddeld
+nr_high: Hoog
+nr_highest: Hoogste
+nr_hint: "Gebruik ruisvermindering alleen als de afbeelding ook daadwerkelijk ruis bevat, anders zal het effect omgekeerd zijn."
+upscaling: Vergroting
+up_none: Geen
+tta: TTA
+tta_best_effort: "n-pass auto"
+tta_rule: "Vertoon de `n-pass auto` optie."
+tta_hint: "TTA verzekerd een betere kwaliteit, maar het limiteerd de grootte van de afbeelding."
+button_convert: Converteer
+button_download: Download
+hints:
+ - "Als je gebruik maakt van FireFox, gebruik dan CTRL + S om een afbeelding op te slaan. De \"Save Image\" \"Afbeelding Opslaan\" optie werkt niet."
+
diff --git a/webgen/templates/index.html.erb b/webgen/templates/index.html.erb
index 9c26526..7cbcc32 100644
--- a/webgen/templates/index.html.erb
+++ b/webgen/templates/index.html.erb
@@ -62,6 +62,14 @@
한국어
+ /
+
+ Nederlands
+
+ /
+
+ Català
+
<%= t[:description] %>
-
waifu2x
+