API tools faq
paste
Advertisement
TheProfosist

waifu2x-caffe_ver1.2.0.1 README English translated by Google

TheProfosist
Jan 6th, 2019
70
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 39.90 KB | None | 0 0
raw download clone embed print report
  1. waifu2x-caffe (for Windows)
  2. ----------
  3.  
  4. Creator: lltcggie
  5.  
  6. This software only converts the conversion function of the image conversion software "[waifu2x] (https://github.com/nagadomi/waifu2x)
  7. It was rewritten using [Caffe] (http://caffe.berkeleyvision.org/) and built for Windows.
  8. It can convert with CPU, but if you use CUDA (or cuDNN) you can convert faster than CPU.
  9.  
  10. GUI supports English, Japanese, Simplified Chinese, Traditional Chinese, Korean, Turkish, Spanish, Russian, and French.
  11.  
  12. You can download the software in the "Downloads" section of the [here releases page] (https://github.com/lltcggie/waifu2x-caffe/releases).
  13.  
  14.  
  15. Required environment
  16. ----------
  17.  
  18. In order to operate this software, at least the following environment is necessary.
  19.  
  20. * OS: Windows Vista or later 64bit (There is no exe for 32bit)
  21. * Memory: Free memory 1 GB or more (depending on the image size to be converted)
  22. * GPU: NVIDIA GPU with Compute Capability 3.0 or higher (not required when converting with CPU)
  23. * Microsoft Visual C ++ 2015 redistributable package Update 3 (x64 version) must be installed (required)
  24. - The above package is [here] (https://www.microsoft.com/en-US/download/details.aspx?id=53587)
  25. - After pressing the `download` button, select `vcredist_x64.exe` and download and install it.
  26. - If you can not find it, try searching for "Visual C ++ 2015 Redistributable Package Update 3".
  27.  
  28. When converting with cuDNN, further
  29.  
  30. * GPU: NVIDIA GPU with Compute Capability 3.0 or higher
  31.  
  32. If you want to know the Compute Capability of your GPU, please check [this page] (https://developer.nvidia.com/cuda-gpus).
  33.  
  34.  
  35. About cuDNN
  36. --------
  37.  
  38. cuDNN is a library for high-speed machine learning which can be used only with NVIDIA GPU.
  39. Although you can convert with CUDA without using cuDNN, using cuDNN has the following advantages.
  40.  
  41. * Depending on the type of GPU used, you can convert images faster
  42. * You can reduce the usage of VRAM (at least at less than half of CUDA, the difference opens as the division size increases)
  43.  
  44. CuDNN has such advantages, but because of license it can not distribute necessary files for operation.
  45. So people who want to use cuDNN download binary for Windows (v 5.1 RC or later) on this page (https://developer.nvidia.com/cuDNN)
  46. Please put "cudnn64_7.dll" in the folder of waifu2x-caffe.
  47. If you insert dll while software is running, please restart the software.
  48. (To download cuDNN you need to register with NVIDIA Developer and register with CUDA Registered Developers.
  49. CUDA Registered Developers probably has a (simple) review, so it does not mean that you can download cuDNN immediately after you register. )
  50.  
  51. Measurement results of processing speed and VRAM usage in the author's environment are as follows.
  52.  
  53. * GPU: GTX 980 Ti
  54. * VRAM: 6 GB
  55. * Processing content: PNG 4ch image with 1000 * 1000 noise removal and enlargement, JPEG noise removal level 1, magnification factor 2.00, TTA mode unused
  56. * Processing time measurement method: Measured average processing time of 10 times in CUI version. However, processing is done twice in advance at first (in order not to include the time required for initialization)
  57. * Calculation method of VRAM usage: (Maximum VRAM used during processing in GUI version) - (VRAM usage after starting GUI version)
  58.  
  59. cuDNN RGB model
  60.  
  61. | Split size | Processing time | VRAM usage (MB) |
  62. |: -------------- |: ------------- |: ------------------- |
  63. | 100 | 00: 00: 03.170 | 278 |
  64. | 125 | 00: 00: 02.745 | 279 |
  65. | 200 | 00: 00: 02.253 | 365 |
  66. | 250 | 00: 00: 02.147 | 446 |
  67. | 500 | 00: 00: 01.982 | 1110 |
  68.  
  69. CUDA RGB model
  70.  
  71. | Split size | Processing time | VRAM usage (MB) |
  72. |: -------------- |: ------------- |: ------------------- |
  73. | 100 | 00: 00: 0 6.192 | 724 |
  74. | 125 | 00: 00: 0 5.504 | 724 |
  75. | 200 | 00: 00: 04.642 | 1556 |
  76. | 250 | 00: 00: 04.436 | 2345 |
  77. | 500 | Unmeasurable | Unmeasurable (6144 or more) |
  78.  
  79. cuDNN UpRGB model
  80.  
  81. | Split size | Processing time | VRAM usage (MB) |
  82. |: -------------- |: ------------- |: ------------------- |
  83. | 100 | 00: 00: 02.831 | 328 |
  84. | 125 | 00: 00: 02.573 | 329 |
  85. | 200 | 00: 00: 02.261 | 461 |
  86. | 250 | 00: 00: 02.150 | 578 |
  87. | 500 | 00: 00: 01.991 | 1554 |
  88.  
  89. CUDA UpRGB model
  90.  
  91. | Split size | Processing time | VRAM usage (MB) |
  92. |: -------------- |: ------------- |: ------------------- |
  93. | 100 | 00: 00: 03.669 | 788 |
  94. | 125 | 00: 00: 03.382 | 787 |
  95. | 200 | 00: 00: 02.965 | 1596 |
  96. | 250 | 00: 00: 02.852 | 2345 |
  97. | 500 | Unmeasurable | Unmeasurable (6144 or more) |
  98.  
  99.  
  100. Usage (GUI version)
  101. --------
  102.  
  103. "Waifu2x - caffe.exe" is GUI software. Double click to start.
  104. Or you can drag and drop files and folders to "waifu2x-caffe.exe" in Windows Explorer and convert with the setting at the last boot.
  105. In that case, depending on the setting, the dialog is automatically closed when conversion is successful.
  106. Also, you can set options on the command line using the GUI.
  107. For details, please read the command line option (common) and command line option (GUI version).
  108.  
  109. After launch, if you drag and drop images or folders into the "Input Path" column, the "Output Path" field will be set automatically.
  110. If you want to change the output destination, please change the "output path" field.
  111.  
  112. You can change the conversion setting according to your preference.
  113.  
  114.  
  115. ## I / O setting
  116. This is a set of setting items related to file input / output.
  117.  
  118. ### "Input Path"
  119. Specify the path of the file you want to convert.
  120. If you specify a folder, convert the file with the "extension to convert within the folder" including the subfolder.
  121. You can specify multiple files and folders by dragging.
  122. In that case, it will be output in the new folder keeping the file and folder structure.
  123. ("(Multi Files)" is displayed in the input pass column.The output folder name is generated from the file and the folder name grasped with the mouse)
  124. When selecting a file by pressing the Browse button, you can select a single file, folder, or multiple files.
  125.  
  126. ### "Output Path"
  127. Specify the path to save the converted image.
  128. If you specify a folder in "Input path", save the converted file in the specified folder (without changing the folder structure). If there is no specified folder, it will be created automatically.
  129.  
  130. ### "Extensions to convert within folders"
  131. When the "input path" is a folder, specify the extension of the image to be converted in the folder.
  132. The default value is `png: jpg: jpeg: tif: tiff: bmp: tga`.
  133. The delimiter is `:`.
  134. Capital letters and lower case letters are not distinguished.
  135. Eg. Png: jpg: jpeg: tif: tiff: bmp: tga
  136.  
  137. ### "Output extension"
  138. Specify the format of the converted image.
  139. Values ​​that can be set for "Output image quality setting" and "Number of output depth bits" differ depending on the format specified here.
  140.  
  141. ### "Output image quality setting"
  142. Specify the image quality of the converted image.
  143. Values ​​that can be set are integers.
  144. The range and meaning of possible values ​​depend on the format set in "Output extension".
  145.  
  146. * .jpg: Range of values ​​(0 to 100) The higher the number, the higher the image quality
  147. * .webp: Range of values ​​(1 to 100) If the number is high and the image quality is 100, lossless compression
  148. *. tga: Value range (0 to 1) 0 means no compression, 1 means RLE compression
  149.  
  150. ### "Number of output depth bits"
  151. Specify the number of bits per channel of the converted image.
  152. Values ​​that can be specified vary depending on the format set in "Output extension".
  153.  
  154. ## Conversion image quality / processing setting
  155. Processing method of file conversion, Setting items related to image quality.
  156.  
  157. ### "Conversion mode"
  158. Specify the conversion mode.
  159. * Noise elimination and expansion: Noise elimination and expansion
  160. * Expansion: We will expand
  161. * Noise removal: Noise removal
  162. * Noise elimination (automatic discrimination) and enlargement: Perform enlargement. It also performs noise removal only when the input is a JPEG image
  163.  
  164. ### "JPEG noise removal level"
  165. Specify the noise removal level. The higher level removes the noise more strongly, but there is also the possibility of becoming a slick picture.
  166.  
  167. ### "Enlarged size"
  168. Set the size after enlargement.
  169. * Specified by enlargement ratio: Enlarges the image at the specified enlargement ratio
  170. * Specified by the width after conversion: While enlarging the aspect ratio of the image, enlarges it so that it becomes the specified width (unit is pixel)
  171. * Specified by the vertical width after conversion: While maintaining the aspect ratio of the image, enlarge it so that it becomes the designated vertical width (unit is pixel)
  172. * Specified by width and width after conversion: Enlarges so that it becomes the specified vertical and horizontal width. Specify it like "1920 x 1080" (unit is pixel)
  173. In the case of magnification ratios exceeding 2 times (in the case of removing noise, only the first one is performed), the enlargement ratio is doubled until exceeding the designated enlargement ratio, and in the case of the enlargement ratio not being a power of 2, We will do the process. Therefore, there is a possibility that the result of conversion will be a solid picture.
  174.  
  175. ### "model"
  176. Specify the model to use.
  177. Since the optimum model differs depending on the image to be converted, we recommend trying various models.
  178. * 2-dimensional illustration (RGB model): 2-dimensional illustration model for converting all RGB of image
  179. * Photo · Anime (Photo model): Model for photography · animation
  180. * 2 dimensional illustration (UpRGB model): A model that transforms at higher speed and equal or higher image quality than 2D illustration (RGB model). However, since the amount of memory (VRAM) consumed by the RGB model is large, adjust the partition size when forcibly terminating during conversion
  181. * Photo · Animation (UpPhoto model): A model that converts with higher image quality than the photo / animation (Photo model) at high speed. However, since the amount of memory (VRAM) consumed by the Photo model is large, adjust the split size when forcibly terminating during conversion
  182. * 2-dimensional illustration (Y model): 2-dimensional illustration model that converts only the luminance of the image
  183. * 2 dimensional illustration (UpResNet 10 model): A model that converts with higher image quality than 2D illustration (UpRGB model). Note that this model changes the output result if the division size is different, so be careful
  184. * 2 Dimensional illustration (CUnet model): A model that can convert 2D illustration with the highest image quality in the bundled model. Note that this model changes the output result if the division size is different, so be careful
  185.  
  186. ### "Using TTA Mode"
  187. Specify whether to use TTA (Test-Time Augmentation) mode or not.
  188. If you use TTA mode, PSNR (image evaluation index) will rise about 0.15 instead of 8 times slower conversion.
  189.  
  190. ## Processing speed setting
  191. It is a set of setting items that affect the processing speed of image conversion.
  192.  
  193. ### "split size"
  194. Specify the width (in pixels) when processing is performed internally.
  195. The best way to decide the numerical value (processing is the fastest) is explained in the section "Split size".
  196. The upper one delimited by "-------" is the divisor of the vertical and horizontal sizes of the input image,
  197. The lower one is a general-purpose division size read from "crop_size_list.txt".
  198. If the partition size is too large, beware that the amount of memory required (the amount of VRAM when GPU is used) exceeds the memory available on the PC and the software terminates forcibly.
  199. Since it affects the processing speed to a certain extent, when converting a large number of images with the same image size by folder specification, we recommend that you convert from the optimal partition size before converting.
  200. However, depending on the model, please note that if you change the partition size, the output result will change.
  201. (In that case, you can increase the processing speed by using the default for the division size and adjusting the batch size)
  202.  
  203. ### "Batch Size"
  204. Specify the size when processing is performed collectively inside.
  205. Increasing the batch size may increase processing speed.
  206. Please be careful that the amount of memory required as well as the partition size does not exceed the memory available on the PC.
  207.  
  208. ## Operation setting
  209. It is a group of settings that summarizes the behavior settings that seems to have little chance to change.
  210.  
  211. ### "Automatic conversion start setting at file input setting"
  212. It sets whether to automatically start converting when specifying an input file with reference button or drag and drop.
  213. In the case where the input file is given as an argument to exe, the setting contents of this item will not be affected.
  214. * Do not start automatically: Even if you enter a file, conversion will not start automatically
  215. * Start with inputting even one file: Start converting automatically when you input even one file
  216. * Start when you input a folder or multiple files: Start folder: When multiple files are input, conversion starts automatically. Please convert a single image file only when adjusting conversion settings
  217.  
  218. ### "Processor Used"
  219. Specify the processor that performs the conversion.
  220. * CUDA (cuDNN if used): Convert using CUDA (GPU) (cuDNN is used if cuDNN is available)
  221. * CPU: Convert using CPU only
  222.  
  223. ### "Do not overwrite output file"
  224. If this setting is ON, if there is a file with the same name in the image writing destination, no conversion will be done.
  225.  
  226. ### "Startup setting with argument"
  227. Sets the behavior when input file is given as argument to exe.
  228. * Convert at startup: Conversion starts automatically at startup
  229. * End on success: Automatically terminate unless conversion failed at the end of conversion
  230.  
  231. ### "GPU No. used"
  232. You can specify the device number to use when there are multiple GPUs. Ignored when in CPU mode or when an invalid device number is specified.
  233.  
  234. ### "Input reference fixed folder"
  235. Fix the folder displayed first when you press the input reference button to the folder set here.
  236.  
  237. ### "Fixed folder when referring to output"
  238. Fix the output destination folder of the converted image to the folder set here.
  239. Also, fix the folder displayed first when you press the output reference button to the folder set here.
  240.  
  241. ## Other
  242. Other setting items group.
  243.  
  244. ### "UI language"
  245. Set the language of the UI.
  246. The same language as the language setting of the PC is selected at the first start. (It will be in English if it does not exist)
  247.  
  248. ### "cuDNN check"
  249. You can check if you can use cuDNN by pressing "cuDNN check" button.
  250. If cuDNN can not be used, the reason is displayed.
  251.  
  252. Click "Execute" button to start conversion.
  253. If you want to cancel halfway, press "Cancel" button.
  254. However, there is a time lag until actually stopping.
  255. The progress bar shows the degree of progress when changing multiple images.
  256. The expected remaining time is displayed in the log, but this is an estimate when processing multiple files of the same vertical width and horizontal width.
  257. So it is not useful when the size of the file falls apart, and only "unknown" is displayed when the number of processed images is 2 or less.
  258.  
  259.  
  260. How to use (CUI version)
  261. --------
  262.  
  263. "Waifu2x - caffe - cui.exe" is a command line tool.
  264. Start `command prompt` and type in the command as follows.
  265.  
  266.  
  267. The following commands output usage on the screen.
  268. `` `
  269. waifu2x-caffe-cui.exe --help
  270. `` `
  271.  
  272. The following commands are examples of commands that perform image conversion.
  273. `` `
  274. waifu2x-caffe-cui.exe -i mywaifu.png -m noise_scale --scale_ratio 1.6 - noise_level 2
  275. `` `
  276. When you do the above, the conversion result is saved in `mywaifu (noise_scale) (Level 2) (x 1.600000) .png`.
  277.  
  278. For details of the command list and each command, please read the command line option (common) and command line option (CUI version).
  279.  
  280.  
  281. Command line option (common)
  282. --------
  283.  
  284. With this software, the following options can be specified.
  285. In the GUI version, when starting with a command line option other than the input file specified, the option file is not saved currently.
  286. Also, options not specified in the GUI version are used at the time of last exit.
  287.  
  288. ### -l <string>, --input_extension_list <string>
  289. When input_file is a folder, specify the extension of the image to be converted in the folder.
  290. The default value is `png: jpg: jpeg: tif: tiff: bmp: tga`.
  291. The delimiter is `:`.
  292. Eg. Png: jpg: jpeg: tif: tiff: bmp: tga
  293.  
  294. ### -e <string>, --output_extention <string>
  295. When input_file is a folder, specify the extension of the output image.
  296. The default value is `png`.
  297.  
  298. ### - m <noise | scale | noise_scale>, --mode <noise | scale | noise_scale>
  299. Specify the conversion mode. If not specified, `noise_scale` is selected.
  300. * noise: Performs noise removal (to be exact, image conversion is performed using a noise removal model)
  301. * scale: Enlarges (exactly, after enlarging with the existing algorithm, image conversion is performed using the model for complementing enlarged image)
  302. * noise_scale: Performs noise elimination and enlargement (enlargement processing continues after removing noise)
  303. * auto_scale: Enlarges. It also performs noise removal only when the input is a JPEG image
  304.  
  305. ### -s <number with decimal point>, --scale_ratio <number with decimal point>
  306. Specify how many times the image is enlarged. The default value is `2.0`, but you can specify anything other than 2.0 times.
  307. If scale_width or scale_height is specified, that will take precedence.
  308. If you specify a number other than 2.0, the following processing is performed.
  309. * First of all, repeat double enlargement so as to cover the specified magnification sufficiently.
  310. If a number other than the power of 2 is specified, the image enlarged so as to be the specified magnification is reduced with a linear filter.
  311.  
  312. ### - w <integer>, --scale_width <integer>
  313. While keeping the aspect ratio of the image, enlarge it to the specified horizontal width (unit: pixels).
  314. When it is specified together with scale_height, the image is enlarged so that it becomes the specified vertical and horizontal width.
  315.  
  316. ### - h <integer>, --scale_height <integer>
  317. While maintaining the aspect ratio of the image, enlarge it to the specified vertical width (unit is pixel).
  318. When it is specified together with scale_width, it enlarges the image so that it becomes the specified vertical and horizontal width.
  319.  
  320. ### - n <0 | 1 | 2 | 3>, - noise_level <0 | 1 | 2 | 3>
  321. Specify the noise removal level. Since only the levels 0 to 3 are prepared for the noise removal model,
  322. Please specify 0, 1, 2 or 3.
  323. The default value is `0`.
  324.  
  325. ### - p <cpu | gpu | cudnn>, - process <cpu | gpu | cudnn>
  326. Specify the processor to use for processing. The default value is `gpu`.
  327. * cpu: Convert using CPU.
  328. * gpu: Convert using CUDA (GPU). For Windows version only use cuDNN if you can use cuDNN.
  329. * cudnn: Convert using cuDNN.
  330.  
  331. ### - c <integer>, --crop_size <integer>
  332. Specify the division size. The default value is `128`.
  333.  
  334. ### - q <integer>, --output_quality <integer>
  335. Set the image quality of the converted image. The default value is `-1`
  336. Specifiable values ​​and meanings depend on the format set in "Output extension".
  337. If -1, default values ​​for each image format are used.
  338.  
  339. ### - d <integer>, --output_depth <integer>
  340. Specify the number of bits per channel of the converted image. The default value is `8`.
  341. Values ​​that can be specified vary depending on the format set in "Output extension".
  342.  
  343. ### - b <integer>, - batch_size <integer>
  344. Specify the mini-batch size. The default value is `1`.
  345. The mini-batch size is the number of simultaneous processing of blocks obtained by dividing an image by "division size". For example, if you specify `2`, it will convert two blocks at a time.
  346. mini-batch Increasing the size increases the GPU usage rate as well as increasing the split size, but if you feel it is measured, it is more effective to increase the split size.
  347. (For example, if you set split size to `128` and mini-batch size to` 1` rather than split size set to `64` and mini-batch size to` 4`, processing will be finished faster)
  348.  
  349. ### - gpu <integer>
  350. Specify the GPU device number to be used for processing. The default value is `0`.
  351. Please note that the GPU device number starts with 0.
  352. Ignored if you do not use GPU for processing.
  353. Also, if you specify a GPU device number that does not exist, it will be executed with the default GPU.
  354.  
  355. ### - t <0 | 1>, - tta <0 | 1>
  356. If `1` is specified, TTA mode is used. The default value is `0`.
  357.  
  358. ### -, --ignore_rest
  359. Ignores all options after this option is specified.
  360. It is for script batch file.
  361.  
  362.  
  363. Command line option (GUI version)
  364. --------
  365.  
  366. In the GUI version, arguments that do not fit option specifications are recognized as input files.
  367. You can specify files, folders, multiple files, and files and folders at the same time as input files.
  368.  
  369. ### -o <string>, --output_folder <string>
  370. Set the path to the folder where the converted image will be saved.
  371. Save the converted file in the specified folder.
  372. The naming convention of the converted file is the same as the output file name automatically determined when setting the input file in the GUI.
  373. If it is not specified, it is saved in the same folder as the first input file.
  374.  
  375. ### --auto_start <0 | 1>
  376. If `1` is specified, conversion is started automatically at startup.
  377.  
  378. ### --auto_exit <0 | 1>
  379. If you specify `1`, it automatically ends when the conversion succeeds if it is automatically converted at startup.
  380.  
  381. ### - no overwrite <0 | 1>
  382. If `1` is specified, if there is a file with the same name in the image writing destination, no conversion will be done.
  383.  
  384. upconv_7_photo | anime_style_art_rgb | photo | anime_style_art_y | upresnet10 | cunet>
  385. Specify the model to use.
  386. It corresponds to the setting item "model" in the GUI as follows.
  387. * upconv_7_anime_style_art_ rgb: 2D illustration (UpRGB model)
  388. * upconv_7_photo: Photography · Anime (UpPhoto model)
  389. * anime_style_art_rgb: 2D illustration (RGB model)
  390. * photo: Photo · Anime (Photo model)
  391. * anime_style_art_y: 2-dimensional illustration (Y model)
  392. * upresnet10: 2D illustration (UpResNet 10 model)
  393. * cunet: 2 dimensional illustration (CUnet model)
  394.  
  395.  
  396. Command line option (CUI version)
  397. --------
  398.  
  399. ### --version
  400. Output version information and exit.
  401.  
  402. ### - ?, --help
  403. Display usage and exit.
  404. Please feel free to check how to use.
  405.  
  406. ### -i <string>, --input_file <string>
  407. (Required) Path to the image to be converted
  408. If you specify a folder, it will convert all image files below that folder and output to the folder specified by output_file.
  409.  
  410. ### -o <string>, --output_file <string>
  411. Path to the file to save the converted image
  412. (When input_file is a folder) Path to the folder where the converted image is saved
  413. (When input_file is an image file) Please be sure to input the extension (last .png etc.).
  414. If not specified, the file name is automatically determined and saved in that file.
  415. The file name determination rule is
  416. `` (Original image file name) `` (model name) `` (mode name) `` (noise removal level (in noise removal mode)) `` (magnification scale in magnified mode) `` Number of bits (except for 8 bits)) `` Output extension `
  417. It has become like.
  418. Basically, the saved location is the same directory as the input image.
  419.  
  420. ### --model_dir <string>
  421. Specify the path to the directory where the model is stored. The default value is `models / cunet`.
  422. The following models are included with the standard.
  423. * `models / anime_style_art_rgb`: 2D illustration (RGB model)
  424. * `models / anime_style_art`: 2D illustration (Y model)
  425. * `models / photo`: Photo · Anime (Photo model)
  426. * `models / upconv_7_anime_style_art_rgb`: 2D illustration (UpRGB model)
  427. * `models / upconv_7_photo`: Photography · Anime (UpPhoto model)
  428. * `models / upresnet10`: 2D illustration (UpResNet 10 model)
  429. * `models / cunet`: 2D illustration (CUnet model)
  430. * `models / ukbench`: old-fashioned model for photographs (Only model to be enlarged is attached, noise can not be removed)
  431. Basically it does not have to be specified. Please specify it when using a model other than the default or your own model.
  432.  
  433. ### --crop_w <integer>
  434. Specify the division size (width). If not set, the value of crop_size is used.
  435. Specifying the divisor of the width of the input image may make it possible to convert faster.
  436.  
  437. ### --crop_h <integer>
  438. Specify the division size (vertical width). If not set, the value of crop_size is used.
  439. Specifying the divisor of the vertical width of the input image may make it possible to convert it faster.
  440.  
  441.  
  442. Split size
  443. --------
  444.  
  445. When waifu 2x-caffe (waifu 2 x is also) converts the image,
  446. We divide the image into pieces of a certain size, convert one by one, and finally combine them to return them to a single image.
  447. The split size (crop_size) is the width (in pixels) when splitting this image.
  448.  
  449. If the GPU is not fully used even when converting with CUDA (the GPU usage rate has not reached 100%),
  450. Increasing this number may cause processing to be completed quickly. (Because it will be possible to use up GPU)
  451. Please adjust while watching GPU Load (GPU usage) and Memory Used (VRAM usage) at [GPU - Z] (http://www.techpowerup.com/gpuz/).
  452. In addition, please refer to it because it has the following characteristics.
  453.  
  454. * The higher the number is, the more it will not become faster
  455. * If the division size is a divisor of the vertical and horizontal sizes of the image (or a small number when it is divided), the amount of computation wastefully decreases and becomes faster. (There seems to be the case that the numerical value that does not fit much with this condition becomes the fastest?
  456. * If you doubled the number, in theory the amount of memory used will be quadrupled (actually 3 to 4 times etc) so be careful not to drop the software. Especially since CUDA consumes more memory than cuDNN, be careful
  457.  
  458.  
  459. About images with alpha channel
  460. --------
  461.  
  462. With this software, expansion of images with alpha channel is also supported.
  463. However, since it is processing to enlarge the alpha channel alone, please note that it takes about twice as long as the image without alpha channel does not expand.
  464. However, when the alpha channel is composed of a single color, it can be enlarged in about the same time as when there is no alpha channel.
  465.  
  466.  
  467. The format of language files
  468. --------
  469.  
  470. Language files format is JSON.
  471. If you create new language file, add language setting to 'lang / LangList.txt'.
  472. 'lang / LangList.txt' format is TSV (Tab-Separated Values).
  473.  
  474. * LangName: Language name
  475. * LangID: Primary language [See MSDN] (https://msdn.microsoft.com/en-us/library/windows/desktop/dd318693.aspx)
  476. * SubLangID: Sublanguage [See MSDN] (https://msdn.microsoft.com/en-us/library/windows/desktop/dd318693.aspx)
  477. * FileName: Language file name
  478.  
  479. ex.
  480.  
  481. * Japanese Lang ID: 0x11 (LANG_JAPANESE), SubLangID: 0x01 (SUBLANG_JAPANESE_JAPAN)
  482. * English (US) LangID: 0x09 (LANG_ENGLISH), SubLangID: 0x01 (SUBLANG_ENGLISH_US)
  483. * English (UK) LangID: 0x09 (LANG_ENGLISH), SubLangID: 0x02 (SUBLANG_ENGLISH_UK)
  484.  
  485.  
  486. Disclaimer
  487. --------------
  488.  
  489. This software is not guaranteed.
  490. Please use it under the judgment of the user.
  491. The creator shall not bear any obligation.
  492.  
  493.  
  494. Acknowledgments
  495. ------
  496. [Ultraist] (https://twitter.com/ultraistter) who produced the original [waifu2x] (https://github.com/nagadomi/waifu2x) and model and released it under the MIT license Mr.,
  497. [Amigo] (https://twitter.com/WL_Amigo) created [waifu2x-converter] (https://github.com/WL-Amigo/waifu2x-converter-cpp) based on the original waifu2x Mr. (I was quite referring to how to write README and LICENSE.txt, how to use OpenCV, etc.)
  498. Created the original model by creating [waifu2x-chainer] (https://github.com/tsurumeso/waifu2x-chainer), and released it under the MIT license [https: // github.com/tsurumeso)
  499. will be grateful to.
  500. Also, @ paul70078 who translated the message into English, @ yonhakcher who translated the message into Chinese (Simplified), @ mzhboy who gave a pull request for translation in Simplified Chinese,
  501. @ Kenin 0726 who translated the message into Korean, @ aarhirin who suggested improvement of Korean translation,
  502. @ Lizardon1995, @ yoonhakcher who translated the message into Chinese (traditional), @ Scharynche who gave a pull request for Turkish translation, @ Serted who gave a pull request for French translation,
  503. Thanks to JYUNYA for providing GUI version icon.
  504.  
  505.  
  506. Change log
  507. --------------
  508. * ver 1.2.0.1
  509. - Fixed a bug that crash when running CUnet model in an environment where cuDNN can not be used
  510. - When using the CUnet model with the GUI, an error is issued when the division size is not a multiple of 4
  511. * ver 1.2.0
  512. - Incorporated the CUnet model from waifu2x v0.13.2
  513. - Fixed that cuDNN version check did not work well
  514. * ver 1.1.9.2
  515. - Fixed a crash crash when running UpResNet 10 model in an environment where cuDNN can not be used
  516. * ver 1.1.9.1
  517. - Fix to fail to read English language file in GUI
  518. - Chinese (Simplified, Traditional) translation update of the GUI
  519. * ver 1.1.9
  520. - imported UpResNet 10 model from waifu 2x-chainer v 1.9.0
  521. - Updated CUDA Toolkit to 10.0.130 (Updated dll also included)
  522. - compatible with cuDNN v7.3.1
  523. - Changed build environment to Visual Studio 2015
  524. - Update caffe
  525. - Allowed to specify batch size also in GUI
  526. * ver 1.1.8.4
  527. - compatible with cuDNN v7
  528. * ver 1.1.8.3
  529. - Fixed that the enlargement of vertical width or horizontal width specification may cause deviation of about 1px from specified size due to calculation error
  530. - Correct that output of CUI may be garbled in non-Japanese environment
  531. - compatible with cuDNN v6
  532. * ver 1.1.8.2
  533. - Fixed a bug that sometimes fails when selecting output folder
  534. - Fixed a bug recognized as horizontal width specification even if vertical width size is specified in GUI
  535. - The output depth bit number can not be changed by input
  536. * ver 1.1.8.1
  537. - Added size specification by width and width after conversion in GUI
  538. - CUI allows you to simultaneously specify width and width after conversion
  539. - French added to translate GUI
  540. * ver 1.1.8
  541. - Updated CUDA Toolkit to 8.0.44 (dll included is also updated)
  542. - cu DNN v 5.1 compatible
  543. - If the alpha channel is monochromatic, speed up by expanding with cv :: INTER_NEAREST
  544. * ver 1.1.7.1
  545. - Add icon to GUI version
  546. - Fixed a bug that output was wrong when using TTA mode when crop_w and crop_h are different values
  547. * ver 1.1.7
  548. - Update upconv model
  549. - Change the model used by the standard to upconv_7_anime_style_art_rgb
  550. - The upconv model is displayed on the GUI version
  551. - Fixed a bug that the output file name does not change even when changing the noise removal level to 0 in the GUI version
  552. * ver 1.1.6.1
  553. - Fixed a bug that denoising level 0 could not be specified with command line option
  554. - Fixed a bug that output file name suffix does not change when noise removal level 0 radio button is pressed in GUI version
  555. - Fixed bug that Chinese (simplified) translation was not displayed correctly
  556. * ver 1.1.6
  557. - Add noise removal level 0
  558. - Change the default value of noise removal level from 1 to 0
  559. - Fixed a bug where vertical width specification enlargement behavior was the same as horizontal width specification
  560. - Fixed that the use amount of VRAM exceeding the maximum value was forcibly terminated in algorithm search using cuDNN
  561. - Fixed a bug that forcibly terminates when selecting a large number of files in GUI input input reference dialog
  562. - Fixed a bug that command line option crop size specification was not working in GUI version
  563. - Correct that strange images appear when converting 32 bit Bitmap
  564. * ver 1.1.5
  565. - Added upconv model (photo)
  566. - Photo Noise Reduction Model Update
  567. - compatible with cuDNN v5.1 RC
  568. - Faster processing speed when using cuDNN
  569. (we made cu DNN usage algorithm fastest and cached the result)
  570. - Added crop_w, crop_h option in CUI version
  571. - Update contents of "About cuDNN" of README according to the current situation
  572. - Messages are issued on forced termination in the GUI version
  573. (Although only message that split size may be the cause is issued, but in fact it may be another cause to note)
  574. - Fixed a bug that forcibly terminates if batch_size is 2 or more
  575. - Fixed a bug that forcibly terminates if the used processor is CPU
  576. - Fixed cuBLAS and cuRAND initialization failure message issued in an environment where CUDA is not available
  577. * ver 1.1.4
  578. - correspond to upconv model (illustration)
  579. - Noise removal model update
  580. - Enabled to specify the GPU device used for conversion
  581. - The rule of automatic generation of output file name of CUI version is aligned with GUI version
  582. - Memory shortage countermeasure became ineffective due to the relationship which rewrote considerably contents
  583. (In environments where 16 GB of memory is loaded, it may become to forcibly terminate if the image of 1500 x 1500 magnified is enlarged more than 8 times.
  584. Conversely speaking, there is no effect unless it is expanded to a considerable size)
  585. * ver 1.1.3.2
  586. - Fixed a bug that input path might not be recognized when launching by passing command line option in GUI
  587. - Update Caffe (corresponds to cuDNN v5 RC)
  588. * ver 1.1.3.1
  589. - Corresponds to command line options with GUI
  590. - Separate option settings in separate windows with GUI
  591. - Added behavior setting when file is specified as argument in GUI
  592. - Added setting to prohibit overwriting of output file in GUI
  593. - Added initial directory setting when pressing reference button in GUI
  594. - Add reference button to output in GUI
  595. - Improved so that you can select files and folders at the same time in the window when input reference button is pressed in GUI
  596. - Fixed a bug that forced termination occurred with 16 bit output
  597. - Added check on whether GUI is Compute Capability 2.0 or higher device
  598. - Reduced memory usage
  599. - If the size of the enlarged image exceeds 3 GB, data is written to a temporary file so as to take measures against memory shortage
  600. * ver 1.1.3
  601. - Implement noise rejection level 3
  602. - Update model of anime_style_art (Y direction)
  603. - Added ability to change the size of the dialog displayed by GUI reference button
  604. - Add translation for Traditional Chinese
  605. - Add translation for Turkish
  606. * ver 1.1.2.1
  607. - Fixed a bug that CUI did not start (change the shortcut option for help display to "-?")
  608. * ver 1.1.2
  609. - Enlarged size can be specified by vertical width or horizontal width after conversion
  610. - The remaining time is displayed in GUI when processing multiple files
  611. - Fixed GUI freezing for a long time when inputting a folder containing a large number of images in the GUI
  612. - Improve image quality for magnification ratios of sizes other than factorial size of 2 (Reduction algorithm changed from Linear to Lanczos 4)
  613. - Supports input of multiple files and folders
  614. - When passing image files and folders as command line arguments to GUI exe, conversion was done at the setting at the last start
  615. * ver 1.1.1.3
  616. - Add translation for Korean
  617. - Fixed that the progress bar was not functioning in the GUI
  618. * ver 1.1.1.2
  619. - Add translation for Simplified Chinese
  620. - Fixed a bug that sometimes the output result of partial image format may be incorrect
  621. - Enabled to set whether to use RLE compression when outputting tga
  622. * ver 1.1.1.1
  623. - Fixed bug that alpha channel disappeared due to noise removal
  624. - Fixed a bug that was not enlarged by "noise removal (automatic discrimination) and enlargement" (auto_scale)
  625. * ver 1.1.1
  626. - Fixed that noise appears at magnification ratio more than twice depending on the model to be used
  627. * ver 1.1.0
  628. - Updated enlarged model of 2D illustration (RGB model)
  629. - Supported English with GUI (If you write even language file you should be able to support other languages ​​too)
  630. - Change GUI design
  631. - Update Caffe
  632. - Update cuDNN that uses Caffe with updating to v4 RC
  633. - Added button to set input path in GUI
  634. - Fixed an issue where drag & drop of files is accepted even if GUI is activated with administrator's privilege
  635. - Update OpenCV to v3.1.0
  636. - Corresponds to input and output of WebP with OpenCV update
  637. - Output image quality can be set with output format corresponding to lossy compression
  638. - It is possible to set the number of depth bits of the output image
  639. - Supports input of 16 bit, 32 bit depth image depth
  640. * ver 1.0.8.1
  641. - Restore conversion settings at last exit when starting with GUI
  642. - Fixed forced termination when trying to convert using GPU depending on Compute Capability
  643. - Logs are generated only on error (generated as "error_log _ ~" in the current directory)
  644. * ver 1.0.8
  645. - Update Caffe
  646. - CuDNN to use with Caffe update also updated to v3
  647. - Updated CUDA Toolkit to 7.5.18 (Updated bundled dll)
  648. - Updated photographic model (you can expand not only noise removal)
  649. - Adjust processing of the alpha channel to that of the original one
  650. (It turned out that blurring near the boundary cured when converting an image with an alpha channel, it took about twice as long to expand the image with an alpha channel)
  651. - The image is rounded off when returning from the floating-point type to the integer type (adjusted to the original family)
  652. * ver 1.0.7
  653. - Update model to waifu2x v0.11
  654. - Implementation of TTA (Test-Time Augmentation) mode
  655. - Input_extension_list compare file extension regardless of uppercase and lowercase
  656. * ver 1.0.6.1
  657. - Add tga to input_extension_list
  658. * ver 1.0.6
  659. - Make tga format readable and writable
  660. - Fixed jaggy appearing in antialiasing when converting image with alpha channel
  661. * ver 1.0.5
  662. - Added RGB version model and photo model
  663. - Updated CUDA Toolkit to 7.0.28 (dll included is also updated)
  664. - Even if you specify how many split sizes, the result will be the same
  665. - Speed ​​up initialization
  666. * ver 1.0.4
  667. - Updated Caffe (Added setting for Compute Capability 5.2)
  668. - I checked version of CUDA driver
  669. - Fixed a bug where the default split size might be very large in the GUI
  670. * ver 1.0.3
  671. - Update Caffe
  672. - Updated CUDA Toolkit to 6.5.19 (Updated dll is also included)
  673. - Slightly speed up memory transfer from GPU to CPU
  674. - crop_size (division size), batch_size (CUI version only) can be specified
  675. - Prevent vu cuDNN before v2 (because there are many bugs)
  676. - The reason is displayed when the GUI cuDNN check fails
  677. - Corrected that "- j" option which does not exist in waifu 2 x - caffe - cui.exe execution example of README.txt was attached.
  678. - Various additions of README.txt
  679. - Because it is misleading, I changed a part of the expression GPU to CUDA
  680. * ver 1.0.2
  681. - Fixed a bug that caused strange output when converting images with alpha channel
  682. - Fixed a bug that output file name does not change even if noise removal level is changed with GUI
  683. - Fixed a bug where cuDNN could not be used if gpu was specified as the conversion processor
  684. - Fixed that the format of the automatic GUI setting output name is different from CUI
  685. - I tried to display the processing time and the processor I used in the GUI
  686. * ver 1.0.1
  687. - Fixed bug that CUI did not work
  688. * ver 1.0.0
  689. - Release
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement
Public Pastes
Advertisement
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!