lltcggie / waifu2x-caffe Goto Github PK

I found my waifu2x-caffe can't use cudnn any more. It will cause system frozen and have to be shut down forcibly. I've tried different combinations to locate the problem.

Case 1:
waifu2x-caffe v1.1.7.1 + cudnn-8.0 v5.1 + GeForce 372.70 / 368.81
System frozen.

Case 2:
waifu2x-caffe v1.1.7.1 + cudnn-8.0 v5.1 + GeForce 364.96
Error: Input can not be converted by GPU. The CUDA driver may not have been installed. Please install the CUDA driver.

Case 3:
waifu2x-caffe v1.1.3.2 + cudnn-7.0 v4.0 + GeForce 364.96
You can not use cuDNN. cFailed to intialize cuDNN.

So what can I do to fix this?

The latest version 1.0.8 always crash (application crash) whenever I press the "run" button no matter parameters I set, the previous version (1.0.7) does not crash
in japanese:その最新ヴァージョン（1.0.8）は、どんなパラメーターに設定しようが run ボタンを押すと必ずクラッシュ（異常終了）してしまいます。以前のヴァージョン（1.0.7）はクラッシュしません。

OS:Win 7 pro 64 bit
btw does waifu2x-caffe support SLI?

So i tried do Magnify a bunch of .bmp files but the output comes out with a messed up alpha channel.


Just wanted to say thanks in advance and keep up the good work!

1.1.5 may request several times of VRAM when algorithm benchmark runs.

This is a log of my VRAM usage. Here I show 6 operation with different crop_size.

Stage 1: crop_size = 576
Stage 2: crop_size = 288
Stage 3: crop_size = 384
Stage 4: crop_size = 576

I close waifu2x-caffe-gui and then open it.
Stage 5: crop_size = 384
Stage 6: crop_size = 576

In Stage 2/3, there is a burst of VRAM usage, but not always happens.

If crop_size = 576 and the usage goes beyong 4GB of my limit, the error log is

Log file created at: 2016/07/08 23:10:26
Running on machine: SERVER_USAGE
Log line format: [IWEF]mmdd hh:mm:ss.uuuuuu threadid file:line] msg
F0708 23:10:26.411075  6272 cudnn_conv_layer.cpp:191] Check failed: status == CUDNN_STATUS_SUCCESS (4 vs. 0)  CUDNN_STATUS_INTERNAL_ERROR

1.1.4 doesn't have this bug.

Hi,

First I would like to thank you guys for creating such a useful tool!

But now we have met a tricky issue.
After using the tool for some period of time on a computer, we will always meet an error where the tool (both GUI and CUI version) will fail to open the input file.

"Failed to open input file xxx"

It occurs on TIFF file format (whether as input or output), but for TGA and possibly other file formats the error does not occur. Currently, only reinstalling the system on the computer can we fix this issue.

We are integrating the tool into an image processing pipeline where TIFF is the intermediate exchange format. Of course we can convert the images to TGA before calling Waifu2x, but I would still like to know the reason why this error occurs, and possibly get it fixed.

Thanks,
Mieu

I chose "简体中文" but showed English.

1.透過PNGに処理できますが、結果がおかしいです。

2.GUIの中でノイズ除去レベルを変えても出力のファイル名は変わらない。

隊列に複数のファイルがある場合、一時停止するオプションが欲しいです。例えば入力フォルダには数万のピクチャーがあって、処理するには百時間以上必要となります。その処理をしている間、他のプログラムはGPUを使えません。（例えばゲームしたいとき。）一時停止ができれば、今処理中のファイルを完成して、次のファイルに読み込まず、他のプログラムがGPUを使い終わったら処理を再開すれば便利かなって考えてます。よろしくお願いします。

Ver1.1.5 GUI版 Windows7（64bit）での動作です。
GUI版で画像の拡大サイズ指定において、縦幅のピクセルを指定して拡大しようとすると、指定したピクセルが横幅のピクセルとなってしまいます。

このように、縦幅に1920ピクセルを指定していますが、実行後に作成されるファイルは縦幅1920ピクセルではなく、横幅1920ピクセルになっていました。

倍率に2.0を指定すると指定通り2倍になりますし、横幅に1920を指定すると、指定通り横幅1920ピクセルの画像が作成されます。
縦幅の指定だけがうまくいっていない様子です。

waifu2x-caffe/waifu2x-caffe-gui/MainDialog.cppでincludeしているtclapw/CmdLine.hがソースツリーに入っていないようです。

Apologies for my poor English, Thanks for your project.
I'm trying to use Waifu2X on M.U.G.E.N chars which was almost jagged.
And there is another tool named 2xScale，under ultra low resolution Waifu2X have better fill but they have more smooth edges.
here is a picture.
http://r5.loli.io/uaeY7j.png
Maybe in further Waifu2X can also has a ultra low resolution mode like that ,of course this is just my personal opinion,thanks for your contribution.

I am using 1.1.5 to test the new RGB model of waifu2x caffe and I notice some differences between RGB and UpRGB
Test image:

(RGB)(noise)(Level1)(tta)(RGB)(scale)(x8.000000)
Looks smooth

vs
(UpRGB)(noise)(Level1)(tta)(UpRGB)(scale)(x8.000000)

Looks like jagged lines is this an error or a feature of UpRGB?

cuda,cuddnともに選択して実行を押すと「waifu2x-caffe.exeは動作を停止しました」となりGPUが使われてる様子もなくそのまま何も起きません。
他のソフトではcuda使えるのでドライバーがダメとかではないらしい。
Optimus(統合グラフィック+GPU)環境なのでそのせいかもしれないです。

This is original images:
original images zip
Output iamges:
output images zip

Ver.1.1.1にて
「ノイズ除去」 「ノイズ除去(自動判別)と拡大」のいずれかを選択して変換した場合にアルファチャンネルが消失するようです

Questions are as follows:
Under 「waifu2x-caffe ver 1.1.7.1」,
Calling cuDNN failed, cannot find 「cudnn64_5.dll」
System:
Windows 7 x64
Graphics Card:
NVIDIA GTX 970
Driver Version:
369.30

I have installed 「CUDA Version 8.0.44」, how to solve, very grateful !

https://developer.nvidia.com/cuda-toolkit

CUDA toolkit 8.0 released.

the vapoursynth tools of waifu2x-caffe already support 8.0.

https://github.com/HomeOfVapourSynthEvolution/VapourSynth-Waifu2x-caffe

will you update it sometime?

1.1.8

Current Nvidia driver version 368.69
Is there a way to update to CUDA 8.0 while maintaining 368.69?

Once I use video card processing，the card will be remove ,system not detecte the video card,the error like this
gtx950m,driver verson 372.90

I'm new to the case, please correct

From

81 "MessagecuDNNOK":"Вы не можете использовать cuDNN",

To

81 "MessagecuDNNOK":"Вы можете использовать cuDNN",

意図せずアルファチャンネルが残っている画像を処理するときの速度を向上したいと思っております。
そこで、対象画像のアルファチャンネルを破棄して処理するオプションを実装していただけないでしょうか。（主にCUI版）
不躾なお願いで恐縮ですが、何卒ご検討ください。

Simplified Chinese - 简体中文

{
"IDC_STATIC_IO_SETTING":"输入和输出设置",
"IDC_STATIC_INPUT_PATH":"输入\r\n(文件或文件夹)",
"IDC_BUTTON_INPUT_REF":"浏览...",
"IDC_STATIC_OUTPUT_PATH":"输出",
"IDC_STATIC_TANS_EXT_LIST":"文件夹内进行转换的格式",
"IDC_STATIC_OUTPUT_EXT":"输出格式",
"IDC_STATIC_OUTPUT_QUALITY":"输出图像质量",
"IDC_STATIC_OUTPUT_DEPTH":"输出色彩深度",
"IDC_STATIC_AUTO_START":"输入目录时的自动转换设置",
"IDC_RADIO_AUTO_START_NONE":"不启用",
"IDC_RADIO_AUTO_START_ONE":"输入单个文件时启用",
"IDC_RADIO_AUTO_START_MULTI":"输入文件夹或多个文件时启用",
"IDC_STATIC_QUALITY_PROCESS_SETTING":"质量和处理设置",
"IDC_STATIC_TRANS_MODE":"转换模式",
"IDC_RADIO_MODE_NOISE_SCALE":"降噪和放大",
"IDC_RADIO_MODE_SCALE":"仅放大",
"IDC_RADIO_MODE_NOISE":"仅降噪",
"IDC_RADIO_AUTO_SCALE":"自动降噪和放大",
"IDC_STATIC_JPEG_NOISE_LEVEL":"降噪等级",
"IDC_RADIONOISE_LEVEL1":"1 级",
"IDC_RADIONOISE_LEVEL2":"2 级",
"IDC_RADIONOISE_LEVEL3":"3 级",
"IDC_STATIC_SCALE_RATE":"放大尺寸",
"IDC_RADIO_SCALE_RATIO":"放大倍率",
"IDC_RADIO_SCALE_WIDTH":"放大后宽度",
"IDC_RADIO_SCALE_HEIGHT":"放大后长度",
"IDC_STATIC_MODEL":"模式",
"IDC_RADIO_MODEL_RGB":"2D 插图 — RGB 模型",
"IDC_RADIO_MODEL_PHOTO":"照片或动画 — Photo 模型",
"IDC_RADIO_MODEL_Y":"2D 插图 — Y 模型",
"IDC_CHECK_TTA":"TTA 模式",
"IDC_STATIC_PROCESS_SPEED_SETTING":"处理速度设置",
"IDC_STATIC_PROCESSOR":"处理器",
"IDC_RADIO_MODE_GPU":"CUDA (cuDNN 优先)",
"IDC_RADIO_MODE_CPU":"CPU",
"IDC_STATIC_CROP_SIZE":"拆分大小",
"IDC_BUTTON_CHECK_CUDNN":"cuDNN 检查",
"IDC_BUTTON_CANCEL":"取消",
"IDC_BUTTON_EXEC":"开始",
"IDC_STATIC_LANG_UI":"界面语言",
"IDC_BUTTON_OUTPUT_REF":"选择",
"IDC_BUTTON_APP_SETTING":"应用设置...",
"IDC_BUTTON_CLEAR_OUTPUT_DIR":"重设输出路径",
"IDC_STATIC_ARG_START":"带参数的启动设置",
"IDC_CHECK_ARG_START_AUTO":"启动时开始转换",
"IDC_CHECK_ARG_START_SUCCESS_FINISH":"成功后自动退出",
"IDC_STATIC_INPUT_DIR_FIX":"默认输入文件夹",
"IDC_STATIC_OUTPUT_DIR_FIX":"默认输出文件夹",
"IDC_CHECK_OUTPUT_NO_OVERWIRITE":"输出文件不覆盖",
"MessageUseProcessorMode":"使用的处理模式: %s",
"MessageProcessTime":"处理时间: %02d:%02d:%02d.%03d",
"MessageInitTime":"初始化时间: %02d:%02d:%02d.%03d",
"MessagecuDNNCheckTime":"cuDNN 检查时间: %02d:%02d:%02d.%03d",
"MessageTitleError":"错误",
"MessageScaleRateCheckError":"放大倍率必须大于 0 的正数",
"MessageScaleWidthCheckError":"放大尺寸必须大于 0 的正数",
"MessageCropSizeCheckError":"拆分大小必须大于 0 的正数",
"MessageInputPathCheckError":"请设置输入路径",
"MessageOutputPathCheckError":"请设置输出路径",
"MessageOutputExtCheckError":"请设置输出格式",
"MessageInputCheckError":"输入或输出路径不存在",
"MessageOutputExtentionCheckError":"请选择输出格式",
"MessageOutputQualityCheckError":"输出图像质量过大",
"MessageOutputDepthCheckError":"输出图像色彩深度数值不正确",
"MessageCudaNotFindError":"使用 GPU 转换失败。\r\n可能是因为没有安装可调用的 CUDA 驱动。\r\n请安装驱动。",
"MessageCudaOldVersionError":"使用 GPU 转换失败。\r\n可能是因为安装的 CUDA 驱动太旧。\r\n请更新驱动。",
"MessageTransSuccess":"转换成功",
"MessageErrorHappen":"发生错误",
"MessageCreateOutDirError":"输出目录\r\n「%s」\r\n创建失败",
"MessageCancelError":"已经取消",
"MessageInvalidParameterError":"参数无效",
"MessageFailedOpenModelFileError":"模型文件无法打开",
"MessageFailedParseModelFileError":"模型文件已损坏",
"MessageFailedConstructModelError":"无法构建模型",
"MessageFailedOpenInputFileError":"无法打开输入文件「%s」",
"MessageFailedOpenOutputFileError":"无法输出文件「%s」",
"MessageFailedProcessCaffeError":"插值处理失败",
"MessageTitleResult":"结果",
"MessagecuDNNOK":"可以使用 cuDNN。",
"MessagecuDNNNotFindError":"调用 cuDNN 失败。\r\n没有发现「%s」。",
"MessagecuDNNOldVersionError":"调用 cuDNN 失败。\r\n「%s」版本太旧。请使用 %s。",
"MessagecuDNNCannotCreateError":"调用 cuDNN 失败。\r\n初始化 cuDNN 失败。",
"MessagecuDNNDefautlError":"调用 cuDNN 失败",
"MessageExtStr":"输入文件扩展名 (%s)",
"MessageAllFileFolder":"所有文件和文件夹 (.)",
"MessageTitleInputDialog":"请选择输入文件或文件夹",
"MessageTimeLeftUnkown":"估计剩余时间: 未知",
"MessageTimeLeft":"估计剩余时间",
"MessageNoOverwrite":"此位置已经包含同名文件: %s",
"MessageCudaOldDeviceError":"CUDA 组件版本太旧。\r\n请使用 rnCompute Capability 2.0 以上版本。",
"OK":"确定",
"Cancel":"取消"
"IDC_STATIC_USE_GPU_NO":"GPU 相关性",
"IDC_RADIO_MODEL_UPCONV_RGB":"2D 插图 — UpRGB 模型",
"IDC_RADIO_MODEL_UPCONV_PHOTO":"照片或动画 — UpPhoto 模型",
"MessageLogFatalError":"发生了致命错误。\r\n分割尺寸可能过大。"
}

When I try to select a JPEG sequence the program automatically shuts down. It crashes when I select between 500-1000 JPEG pictures.

Not sure why but sometimes the transparency is not kept in the result. I'm not sure exactly what causes it, but it happens for some images and not others.
http://imgur.com/a/7ppIu

Is there a reason this happens that I am not aware of?

issueではありませんが､WebPフォマットの入力対応をお願いします｡
https://developers.google.com/speed/webp/docs/api

ちなみにcudnnv4も出ました

また､本家の2x model､ default parameterも12月11日に更新されました｡

次のバージョンのwaifu2x-caffeを気を長くしてお待ちしております｡

When the time left is unknown it's printed as

"Time Left: Unkown" just missing an N, no big deal

Hello it's me again. I tried using the latest version and compare it to version 1.0.8.1
Test image:

Now the problem is:

All parameters used: 5x scale,RGB, TTA, level 1 denoise, output: PNG
Other Model parameters ver 1.1.0
Photo model:

Y-model:

http://puu.sh/oyUnz.png using this image on default (16bit changed & png) crashes the program (using CPU)

NVIDIA CUDA

???

Current version can only be run on X64 Windows OS, can we add support for X86(32-bit) systems?

I think, you need to add an option to replace original image file with edited image.
Like a special checkbox in settings for that. Because there are many situations when you just need to denoise the image, but not creating another one with even longer name -___-

because other people might want to work on it, too.

about someone have 2 or more gpus working at the same time ,I think waifu2x-caffe can work this way.
The image is divided into two parts. and gpus work separately. last waifu2x-caffe combine the 2 parts images and output.

Current distribution contains GUI and console versions. But there is a DLL project in sources.
I have only VS 2015, which means I can't build caffe and waifu2x-caffe without installing the older 2013 one.
Could you please add DLL-version in your binary distribution?

While using waifu2x-caffe (magnifying, denoising etc.) I noticed that memory clock is only reaching at 3005 MHz (1502.5 MHz). My GPU has a default memory clock of 1753MHz so it should reach (1753*2) 3506MHz on the sensors. I tried other benchmark programs like Heaven or Kombustor to test if the GPU is the problem but the memory clock reached 3506MHz.
Here is a snapshot of the reading on the memory clock while using waifu2x-caffe v1.1.3.2
(GPU-Z)

I also used NVIDIA inspector and while using waifu2x-caffe it only reaches Power state 'P-2'
In theory GPU Power state while using waifu2x-caffe should reach Power state 'P-0'
Reference
P0/P1 - Maximum 3D performance
P2/P3 - Balanced 3D performance-power
http://docs.nvidia.com/gameworks/content/gameworkslibrary/coresdk/nvapi/group__gpupstate.html
https://devtalk.nvidia.com/default/topic/892842/one-weird-trick-to-get-a-maxwell-v2-gpu-to-reach-its-max-memory-clock-/
https://www.einsteinathome.org/forum_thread.php?id=11044

It seems that the prototxt has already contained something useful for training. There still exists, however, some work needed to be done to train another model using native caffe interface.
I would appreciate it if the project would provide some proto texts essential for training and even scripts. Some basic hints are also fine.

使い方を習うため、READMEを読ませていただきました。
日本語の勉強も兼ねて翻訳しました。これは使わなくても大丈夫です。
README for Korean.txt
ところで、おかしなところがいくつかありました。
まず、「拡大サイズ」のところなんですが、

• 変換後の縦幅で指定 : 画像の縦横比を維持したまま、指定された横幅になるように拡大します(単位はピクセル)
  になっております。「縦幅になるように拡大します。」ではないでしょうか？
  それと、その他の説明が処理速度設定の説明と同じです。
  最後に、CUI版の使い方に
  -s <小数点付き数値>, --scale_ratio <小数点付き数値>・・・のところは既に修正なさいましたね。（汗）

README.txtのエンコーディングがANSIになっておりますが、OSの言語設定によっては文字がまともに出ない場合があります。特にANSIにする理由がなければUTF-8をお勧めします。

そして、ver 1.1.2.1の韓国語訳です。
korean.zip

それと、出力拡張子がjpgの時、出力画質設定の値を指定しても保存されず、次の起動の時はまた95に固定されているんですが、これは変えられないんですか？

LangList.txt

;LangName   LangID  SubLangID   FileName
English(US) 0x09    0x01    english.json
日本語   0x11    0x01    japanese.json
简体中文    0x04    0x04    SChinese.json

SChinese.json

{
"IDC_STATIC_IO_SETTING":"输入和输出设置",
"IDC_STATIC_INPUT_PATH":"输入\r\n(文件或文件夹)",
"IDC_BUTTON_INPUT_REF":"选择",
"IDC_STATIC_OUTPUT_PATH":"输出",
"IDC_STATIC_TANS_EXT_LIST":"文件夹内进行转换的格式",
"IDC_STATIC_OUTPUT_EXT":"输出格式",
"IDC_STATIC_OUTPUT_QUALITY":"输出图像质量",
"IDC_STATIC_OUTPUT_DEPTH":"输出色彩深度",
"IDC_STATIC_QUALITY_PROCESS_SETTING":"转换的质量和处理设置",
"IDC_STATIC_TRANS_MODE":"转换模式",
"IDC_RADIO_MODE_NOISE_SCALE":"降噪和放大",
"IDC_RADIO_MODE_SCALE":"仅放大",
"IDC_RADIO_MODE_NOISE":"仅降噪",
"IDC_RADIO_AUTO_SCALE":"自动降噪和放大",
"IDC_STATIC_JPEG_NOISE_LEVEL":"JPEG 降噪等级",
"IDC_RADIONOISE_LEVEL1":"1 级",
"IDC_RADIONOISE_LEVEL2":"2 级",
"IDC_STATIC_SCALE_RATE":"放大率",
"IDC_STATIC_MODEL":"模式",
"IDC_RADIO_MODEL_RGB":"2D 插图 (RGB 模型)",
"IDC_RADIO_MODEL_PHOTO":"照片 或 动画 (Photo 模型)",
"IDC_RADIO_MODEL_Y":"2D 插图 2 (Y 模型)",
"IDC_CHECK_TTA":"TTA 模式",
"IDC_STATIC_PROCESS_SPEED_SETTING":"处理速度设置",
"IDC_STATIC_PROCESSOR":"处理器",
"IDC_RADIO_MODE_GPU":"CUDA (cuDNN 优先)",
"IDC_RADIO_MODE_CPU":"CPU",
"IDC_STATIC_CROP_SIZE":"拆分大小",
"IDC_BUTTON_CHECK_CUDNN":"cuDNN 检查",
"IDC_BUTTON_CANCEL":"取消",
"IDC_BUTTON_EXEC":"开始",
"IDC_STATIC_LANG_UI":"界面语言",
"MessageUseProcessorMode":"使用的处理模式: %s",
"MessageProcessTime":"处理时间: %02d:%02d:%02d.%03d",
"MessageInitTime":"初始化时间: %02d:%02d:%02d.%03d",
"MessagecuDNNCheckTime":"cuDNN 检查时间: %02d:%02d:%02d.%03d",
"MessageTitleError":"错误",
"MessageScaleRateCheckError":"放大倍数必须大于 0.0 的正数",
"MessageCropSizeCheckError":"拆分大小必须大于 0 的正数",
"MessageInputPathCheckError":"请设置输入路径",
"MessageOutputPathCheckError":"请设置输出路径",
"MessageOutputExtCheckError":"请设置输出格式",
"MessageInputCheckError":"输入或输出路径不存在",
"MessageOutputExtentionCheckError":"请选择输出格式",
"MessageOutputQualityCheckError":"输出图像质量过大",
"MessageOutputDepthCheckError":"输出图像色彩深度数值不正确",
"MessageCudaNotFindError":"使用 GPU 转换失败。\r\n可能是因为没有安装可调用的 CUDA 驱动。\r\n请安装驱动。",
"MessageCudaOldVersionError":"使用 GPU 转换失败。\r\n可能是因为安装的 CUDA 驱动太旧。\r\n请更新驱动。",
"MessageTransSuccess":"转换成功",
"MessageErrorHappen":"发生错误",
"MessageCreateOutDirError":"输出目录\r\n「%s」\r\n创建失败",
"MessageCancelError":"已经取消",
"MessageInvalidParameterError":"参数无效",
"MessageFailedOpenModelFileError":"模型文件无法打开",
"MessageFailedParseModelFileError":"模型文件已损坏",
"MessageFailedConstructModelError":"无法构建模型",
"MessageFailedOpenInputFileError":"无法打开输入文件「%s」",
"MessageFailedOpenOutputFileError":"无法输出文件「%s」",
"MessageFailedProcessCaffeError":"插值处理失败",
"MessageTitleResult":"结果",
"MessagecuDNNOK":"可以使用 cuDNN。",
"MessagecuDNNNotFindError":"调用 cuDNN 失败。\r\n没有发现「%s」。",
"MessagecuDNNOldVersionError":"调用 cuDNN 失败。\r\n「%s」版本太旧。请使用 %s。",
"MessagecuDNNCannotCreateError":"调用 cuDNN 失败。\r\n初始化 cuDNN 失败。",
"MessagecuDNNDefautlError":"调用 cuDNN 失败",
"MessageExtStr":"输入文件扩展名 (%s)",
"MessageAllFileFolder":"所有文件和文件夹 (*.*)",
"MessageTitleInputDialog":"请选择输入文件或文件夹"
}

例えば小説とかのスキャン画像をアップスケールする時、今のモデルより画質いい変換可能のモデルは存在するでしょうか。もし存在するなら、追加してほしいです。

Source (640x360)

waifu2x-caffe-cui.exe --tta 0 --crop_w 160 --crop_h 180 -p cudnn -m scale -i foo.png

waifu2x-caffe-cui.exe --tta 1 --crop_w 160 --crop_h 180 -p cudnn -m scale -i foo.png

The problem also doesn't occur when you assign the same value to both crop_w and crop_h.

単語の誤訳および正書法にあわない言葉を正しく書き直しました。
korean.zip

I'm not sure if this is an issue or not, but a guy introduced me this program and it works really great. However, the program only uses half of my CPU, an I7 4770k. He told me that the program constantly uses 100% of his CPU (i5-3570K). So, problem with Hyper Threading? I don't know, just submit the problem here.

*最近気づきましたが､UpConvを使うと､同じ画像で同じ設定で出力しても毎回の出力の容量が違います｡
これは本家による仕様でしょうか｡

*あと本家のUpcovにはnoise modelも入ってるんですが､waifu-caffeだとnoise_scaleだけが入ってます｡これでUpconv モデルでノイズ除去だけする時modelをnoiseに指定してもnoise_scale modelにならないですか?

*ついでにお知らせとアドバイスです
→本家のupconvモデル三日前に更新された模様｡
→本家の方のようにupcovモデルをデフォルトにするようにおすすめします､平均的に良くなるし､速くもなるので
→modelフォルダ階層を本家に合わせて頂けませんか､または更新のログも､ちょっと照合し辛いです｡

つまりこのように:

┬ upconv_7
├art
└photo

┬ vgg_7
├art
├art_y
├photo
└ukbench

*CUIのコマンドはもっと洗練にするはできるでしょうか
例えば--tta <1|0>を→-t を入れるだけでttaモードになる
modelの指定も→
--model_root upconv_7(デフォルトでupcov) --model art
のはいかがでしょうか

よろしくお願いします

hello,
The software is really good, The waifu2x-caffe(ver1.1.1.1) was working fine here before, and today after I updated the VGA driver, waifu2x-caffe.exe didn't work anymore.
The situation was: I double clicked the waifu2x-caffe.exe, no windows appeared but still created a process called "waifu2x-caffe.exe" in Task Manager.
Furthermore, that process was unable to kill in Task Manager. It was unable to be killed even I use software like lockhunter, killbox etc. And since so, I can't shutdown or reboot my PC because the system was unable to kill it and stuck in the shutting down screen. Cut the power was the onlyway.
I tried to delete the folder, change the path, recover my VGA driver but they didn't work.
Here's my device info
OS:Win7 X64
VGA:GTX-660 (driver verson 361.75 now)
I was hopeless and wish i may got some help here,thank you!

There is a very small probability that you will find that your output image (like .png) might be broken. This happened on my machine and my friend's machine.

The broken image has the corrected file size (this might indicate the calculation process is corrected). The first part of image (basically the top lines of a image) is showing correctly, and the rest of it is broken (like displaying black area or error texture).

The probability is very low, like several errors in 100,000 example. But I found they are more likely to occur within a small period of time, or even continues images.

Hi lltcggie,

Could you please add an option in the interface so that the user can specify which device to use if he has more than one GPU in the system? We use this library for video upscaling, so if the processing can be distributed to different GPUs at the same time, the time needed will reduce a lot. Thank you very much.

E0706 15:34:09.373869 10996 common.cpp:126] Cannot create Curand generator. Curand won't be available.

You just drag'n'drop waifu2x-caffe.exe to Command Prompt and hit Enter.
You can create a shortcut: right-click any Windows Explorer window, select New, select Shortcut and input this target
C:\Windows\System32\cmd.exe /c "C:\Program Files\waifu2x-caffe\waifu2x-caffe.exe"
You may need to edit path depending on the location of waifu2x-caffe.exe